GTest Type and Value Parameterized Tests

GTest exposes clean interfaces for parameterizing your tests by value and by type - but what if you want both?

These views do not in any way represent those of NVIDIA or any other organization or institution that I am professionally associated with. These views are entirely my own.

Getting Started

The need for this arose when converting some of Flang’s unit tests to use GTest to conform with the rest of LLVM’s testing framework (Flang is LLVM’s fortran 2018 compiler). Linked here is the revision where we needed to parameterize both by value and type.

Let’s start with an interface to test:

// addone.hpp
#pragma once

template<typename T>
auto addOne(T t) {
  return t + 1;
}

Our first test might look like this:

#include <gtest/gtest.h>
#include "addone.hpp"

TEST(AddOneTests, doAddTo5) {
  ASSERT_EQ(addOne(5), 6) << "addOne(5) != 6!";
}

What if we want to test multiple values without repeating code? Instead of something like this:

TEST(AddOneTests, doAddTo5) {
  ASSERT_EQ(addOne(5), 6) << "addOne(5) != 6!";
  ASSERT_EQ(addOne(6), 7) << "addOne(6) != 7!";
  ...
}

Value Parameterized Tests

We can parameterize our test by value with a fixture:

struct AddOneTestsFixture
    : public testing::TestWithParam<std::tuple<int, int>> {};

TEST_P(AddOneTestsFixture, doAdd) {
  int input = std::get<0>(GetParam());
  int expect = std::get<1>(GetParam());
  ASSERT_EQ(addOne(input), expect)
      << "addOne(" << input << ") != " << expect << "!";
}

INSTANTIATE_TEST_SUITE_P(
    AddOneTests,
    AddOneTestsFixture,
    testing::Values(
      std::make_tuple(1, 2),
      std::make_tuple(3, 4),
      std::make_tuple(9, 10)));

This way, our tests run over all values we pass in at the end:

$ ./tests
Running main() from /tmp/googletest-20201214-81667-fx54ix/googletest-release-1.10.0/googletest/src/gtest_main.cc
[==========] Running 3 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 3 tests from AddOneTests/AddOneTestsFixture
[ RUN      ] AddOneTests/AddOneTestsFixture.doAdd/0
[       OK ] AddOneTests/AddOneTestsFixture.doAdd/0 (0 ms)
[ RUN      ] AddOneTests/AddOneTestsFixture.doAdd/1
[       OK ] AddOneTests/AddOneTestsFixture.doAdd/1 (0 ms)
[ RUN      ] AddOneTests/AddOneTestsFixture.doAdd/2
[       OK ] AddOneTests/AddOneTestsFixture.doAdd/2 (0 ms)
[----------] 3 tests from AddOneTests/AddOneTestsFixture (0 ms total)

[----------] Global test environment tear-down
[==========] 3 tests from 1 test suite ran. (0 ms total)
[  PASSED  ] 3 tests.

Type Parameterized Tests

Our interface addOne takes a template parameter - what if we want to test this on multiple types?

First, we’ll want our fixture to take template parameters and we’ll have to declare the fixture as templated in GTest:

template<typename T>
struct AddOneTestsFixture : public ::testing::Test {};
TYPED_TEST_SUITE_P(AddOneTestsFixture);

And keep the first iteration of our test, but this time using the TypeParam type exposed by the GTest TYPED_TEST_SUITE api:

TYPED_TEST_P(AddOneTestsFixture, doAddOne) {
  ASSERT_EQ(addOne<TypeParam>(5), 6) << "addOne(5) != 6!";
}

We’ll also have to register each test with our typed test suite:

REGISTER_TYPED_TEST_SUITE_P(AddOneTestsFixture, doAddOne);

If we had more tests, you would register them in the same statement as above:

REGISTER_TYPED_TEST_SUITE_P(AddOneTestsFixture, doAddOne, doAddTwo, ...);

We are then able to instantiate our templated test suite with all the types we intend to use with our test suite:

using Types = testing::Types<int, long long, std::size_t>;
INSTANTIATE_TYPED_TEST_SUITE_P(TestPrefix, AddOneTestsFixture, Types);

And our type-parameterized tests are working!

$ ./tests
Running main() from /tmp/googletest-20201214-81667-fx54ix/googletest-release-1.10.0/googletest/src/gtest_main.cc
[==========] Running 4 tests from 4 test suites.
[----------] Global test environment set-up.
[----------] 1 test from TestPrefix/AddOneTestsFixture/0, where TypeParam = int
[ RUN      ] TestPrefix/AddOneTestsFixture/0.doAddOne
[       OK ] TestPrefix/AddOneTestsFixture/0.doAddOne (0 ms)
[----------] 1 test from TestPrefix/AddOneTestsFixture/0 (0 ms total)

[----------] 1 test from TestPrefix/AddOneTestsFixture/2, where TypeParam = long long
[ RUN      ] TestPrefix/AddOneTestsFixture/2.doAddOne
[       OK ] TestPrefix/AddOneTestsFixture/2.doAddOne (0 ms)
[----------] 1 test from TestPrefix/AddOneTestsFixture/2 (0 ms total)

[----------] 1 test from TestPrefix/AddOneTestsFixture/3, where TypeParam = unsigned long
[ RUN      ] TestPrefix/AddOneTestsFixture/3.doAddOne
[       OK ] TestPrefix/AddOneTestsFixture/3.doAddOne (0 ms)
[----------] 1 test from TestPrefix/AddOneTestsFixture/3 (0 ms total)

[----------] Global test environment tear-down
[==========] 4 tests from 4 test suites ran. (0 ms total)
[  PASSED  ] 4 tests.

Type and Value Parameterized Tests

Now is the tricky part - GTest doesn’t expose an API for parameterizing tests over values and types so we have to do some work ourselves.

First, let’s define the types and input data we’ll be parameterizing our tests over:

template <typename T>
using ParamT = std::vector<std::tuple<T, T>>;

static std::tuple<ParamT<int>, ParamT<long long>, ParamT<std::size_t>> allParams{
  { // Test cases for int
    std::make_tuple(1, 2),
    std::make_tuple(5, 6),
    std::make_tuple(9, 10),
  },
  { // Test cases for long long
    std::make_tuple(1, 2),
    std::make_tuple(5, 6),
    std::make_tuple(9, 10),
  },
  { // Test cases for size_t
    std::make_tuple(1, 2),
    std::make_tuple(5, 6),
    std::make_tuple(9, 10),
  },
};

This structure assumes you may want to add test inputs later on that may be different depending on the type. If this is not the case and you know your make_tuple calls are static_cast-able into your parameter type, you may do the following to reduce code duplication:

#define ADDONE_TESTPARAMS                                                      \
  { std::make_tuple(1, 2), std::make_tuple(5, 6), std::make_tuple(9, 10), }
static std::tuple<ParamT<int>, ParamT<long long>, ParamT<std::size_t>> allParams{
        ADDONE_TESTPARAMS, ADDONE_TESTPARAMS, ADDONE_TESTPARAMS, };

Now, let’s refactor our fixture to take the types and values we just defined:

template <typename T>
struct AddOneTestsFixture : public testing::Test {
  AddOneTestsFixture() : params{std::get<ParamT<T>>(allParams)} {}
  ParamT<T> params;
};

You may notice we set params to std::get< ParamT<T> >(allParams) - this is how we accomplish type and value parameterized tests. We use the infrastructure of a type parameterized test, and leverage std::tuple to do the value parameterization.

For the actual test code, we again reuse most of the test from our first type- parameterized test, this time using the params field of our test fixture:

TYPED_TEST_P(AddOneTestsFixture, doAddOne) {

  // Iterate over the parameters configred by our fixture
  for(auto const& [input, expect] : this->params) {

    // The assertions stay the same as in our original type-parameterized test
    ASSERT_EQ(addOne(input), expect)
      << "addOne(" << input << ") != " << expect << "!";
  }
}

And voilà! our tests are parameterized over values and types:

$ ./tests
Running main() from /tmp/googletest-20201214-81667-fx54ix/googletest-release-1.10.0/googletest/src/gtest_main.cc
[==========] Running 3 tests from 3 test suites.
[----------] Global test environment set-up.
[----------] 1 test from TestPrefix/AddOneTestsFixture/0, where TypeParam = int
[ RUN      ] TestPrefix/AddOneTestsFixture/0.doAddOne
[       OK ] TestPrefix/AddOneTestsFixture/0.doAddOne (0 ms)
[----------] 1 test from TestPrefix/AddOneTestsFixture/0 (0 ms total)

[----------] 1 test from TestPrefix/AddOneTestsFixture/1, where TypeParam = long long
[ RUN      ] TestPrefix/AddOneTestsFixture/1.doAddOne
[       OK ] TestPrefix/AddOneTestsFixture/1.doAddOne (0 ms)
[----------] 1 test from TestPrefix/AddOneTestsFixture/1 (0 ms total)

[----------] 1 test from TestPrefix/AddOneTestsFixture/2, where TypeParam = unsigned long
[ RUN      ] TestPrefix/AddOneTestsFixture/2.doAddOne
[       OK ] TestPrefix/AddOneTestsFixture/2.doAddOne (0 ms)
[----------] 1 test from TestPrefix/AddOneTestsFixture/2 (0 ms total)

[----------] Global test environment tear-down
[==========] 3 tests from 3 test suites ran. (0 ms total)
[  PASSED  ] 3 tests.

Full Code Listing

// addone.hpp
#pragma once
template<typename T>
auto addOne(T t) {
  return t + 1;
}
// addone_test.cpp
#include "addone.hpp"
#include <gtest/gtest.h>
#include <tuple>
#include <vector>

template <typename T>
using ParamT = std::vector<std::tuple<T, T>>;

static std::tuple<ParamT<int>, ParamT<long long>, ParamT<std::size_t>> allParams{
  {
    // Test cases for int
    std::make_tuple(1, 2),
    std::make_tuple(5, 6),
    std::make_tuple(9, 10),
  },
  {
    // Test cases for long long
    std::make_tuple(1, 2),
    std::make_tuple(5, 6),
    std::make_tuple(9, 10),
  },
  {
    // Test cases for size_t
    std::make_tuple(1, 2),
    std::make_tuple(5, 6),
    std::make_tuple(9, 10),
  },
};

template <typename T>
struct AddOneTestsFixture : public testing::Test {
  AddOneTestsFixture() : params{std::get<ParamT<T>>(allParams)} {}
  ParamT<T> params;
};

TYPED_TEST_SUITE_P(AddOneTestsFixture);

TYPED_TEST_P(AddOneTestsFixture, doAddOne) {
  for(auto const& [input, expect] : this->params) {
    ASSERT_EQ(addOne(input), expect)
      << "addOne(" << input << ") != " << expect << "!";
  }
}

REGISTER_TYPED_TEST_SUITE_P(AddOneTestsFixture, doAddOne);

using Types = testing::Types<int, long long, std::size_t>;
INSTANTIATE_TYPED_TEST_SUITE_P(TestPrefix, AddOneTestsFixture, Types);

Makefile used for this post:

CFLAGS := -I/usr/local/Cellar/googletest/1.10.0/include
CFLAGS += -L/usr/local/Cellar/googletest/1.10.0/lib -lgtest -lgtest_main
CFLAGS += -lpthread -std=c++17
CXX    =  clang++

all:
	$(CXX) addone_test.cpp $(CFLAGS) -o tests

Contact

You can reach me at any of the links below:

These views do not in any way represent those of NVIDIA or any other organization or institution that I am professionally associated with. These views are entirely my own.

References


Written on Mar 7th, 2021