diff options
Diffstat (limited to 'src/test/Vector2Test.cpp')
-rw-r--r-- | src/test/Vector2Test.cpp | 384 |
1 files changed, 384 insertions, 0 deletions
diff --git a/src/test/Vector2Test.cpp b/src/test/Vector2Test.cpp new file mode 100644 index 0000000..17bca41 --- /dev/null +++ b/src/test/Vector2Test.cpp @@ -0,0 +1,384 @@ +#include <gtest/gtest.h> + +#include <crepe/api/Vector2.h> + +using namespace crepe; + +class Vector2Test : public ::testing::Test { +public: + Vector2<int> int_vec1; + Vector2<int> int_vec2; + Vector2<double> double_vec1; + Vector2<double> double_vec2; + Vector2<long> long_vec1; + Vector2<long> long_vec2; + Vector2<float> float_vec1; + Vector2<float> float_vec2; + + void SetUp() override { + int_vec1 = {1, 2}; + int_vec2 = {3, 4}; + double_vec1 = {1.0, 2.0}; + double_vec2 = {3.0, 4.0}; + long_vec1 = {1, 2}; + long_vec2 = {3, 4}; + float_vec1 = {1.0f, 2.0f}; + float_vec2 = {3.0f, 4.0f}; + } +}; + +TEST_F(Vector2Test, Subtract) { + Vector2<int> result = int_vec1 - int_vec2; + EXPECT_EQ(result.x, -2); + EXPECT_EQ(result.y, -2); + + Vector2<double> result2 = double_vec1 - double_vec2; + EXPECT_FLOAT_EQ(result2.x, -2.0); + EXPECT_FLOAT_EQ(result2.y, -2.0); + + Vector2<long> result3 = long_vec1 - long_vec2; + EXPECT_EQ(result3.x, -2); + EXPECT_EQ(result3.y, -2); + + Vector2<float> result4 = float_vec1 - float_vec2; + EXPECT_FLOAT_EQ(result4.x, -2.0f); + EXPECT_FLOAT_EQ(result4.y, -2.0f); +} + +TEST_F(Vector2Test, SubtractScalar) { + Vector2<int> result = int_vec1 - 1; + EXPECT_EQ(result.x, 0); + EXPECT_EQ(result.y, 1); + + Vector2<double> result2 = double_vec1 - 1.0; + EXPECT_FLOAT_EQ(result2.x, 0.0); + EXPECT_FLOAT_EQ(result2.y, 1.0); + + Vector2<long> result3 = long_vec1 - 1; + EXPECT_EQ(result3.x, 0); + EXPECT_EQ(result3.y, 1); + + Vector2<float> result4 = float_vec1 - 1.0f; + EXPECT_FLOAT_EQ(result4.x, 0.0f); + EXPECT_FLOAT_EQ(result4.y, 1.0f); +} + +TEST_F(Vector2Test, Add) { + Vector2<int> result = int_vec1 + int_vec2; + EXPECT_EQ(result.x, 4); + EXPECT_EQ(result.y, 6); + + Vector2<double> result2 = double_vec1 + double_vec2; + EXPECT_FLOAT_EQ(result2.x, 4.0); + EXPECT_FLOAT_EQ(result2.y, 6.0); + + Vector2<long> result3 = long_vec1 + long_vec2; + EXPECT_EQ(result3.x, 4); + EXPECT_EQ(result3.y, 6); + + Vector2<float> result4 = float_vec1 + float_vec2; + EXPECT_FLOAT_EQ(result4.x, 4.0f); + EXPECT_FLOAT_EQ(result4.y, 6.0f); +} + +TEST_F(Vector2Test, AddScalar) { + Vector2<int> result = int_vec1 + 1; + EXPECT_EQ(result.x, 2); + EXPECT_EQ(result.y, 3); + + Vector2<double> result2 = double_vec1 + 1.0; + EXPECT_FLOAT_EQ(result2.x, 2.0); + EXPECT_FLOAT_EQ(result2.y, 3.0); + + Vector2<long> result3 = long_vec1 + 1; + EXPECT_EQ(result3.x, 2); + EXPECT_EQ(result3.y, 3); + + Vector2<float> result4 = float_vec1 + 1.0f; + EXPECT_FLOAT_EQ(result4.x, 2.0f); + EXPECT_FLOAT_EQ(result4.y, 3.0f); +} + +TEST_F(Vector2Test, Multiply) { + Vector2<int> result = int_vec1 * int_vec2; + EXPECT_EQ(result.x, 3); + EXPECT_EQ(result.y, 8); + + Vector2<double> result2 = double_vec1 * double_vec2; + EXPECT_FLOAT_EQ(result2.x, 3.0); + EXPECT_FLOAT_EQ(result2.y, 8.0); + + Vector2<long> result3 = long_vec1 * long_vec2; + EXPECT_EQ(result3.x, 3); + EXPECT_EQ(result3.y, 8); + + Vector2<float> result4 = float_vec1 * float_vec2; + EXPECT_FLOAT_EQ(result4.x, 3.0f); + EXPECT_FLOAT_EQ(result4.y, 8.0f); +} + +TEST_F(Vector2Test, MultiplyScalar) { + Vector2<int> result = int_vec1 * 2; + EXPECT_EQ(result.x, 2); + EXPECT_EQ(result.y, 4); + + Vector2<double> result2 = double_vec1 * 2.0; + EXPECT_FLOAT_EQ(result2.x, 2.0); + EXPECT_FLOAT_EQ(result2.y, 4.0); + + Vector2<long> result3 = long_vec1 * 2; + EXPECT_EQ(result3.x, 2); + EXPECT_EQ(result3.y, 4); + + Vector2<float> result4 = float_vec1 * 2.0f; + EXPECT_FLOAT_EQ(result4.x, 2.0f); + EXPECT_FLOAT_EQ(result4.y, 4.0f); +} + +TEST_F(Vector2Test, Divide) { + Vector2<int> result = int_vec1 / int_vec2; + EXPECT_EQ(result.x, 0); + EXPECT_EQ(result.y, 0); + + Vector2<double> result2 = double_vec1 / double_vec2; + EXPECT_FLOAT_EQ(result2.x, 0.33333333333333331); + EXPECT_FLOAT_EQ(result2.y, 0.5); + + Vector2<long> result3 = long_vec1 / long_vec2; + EXPECT_EQ(result3.x, 0); + EXPECT_EQ(result3.y, 0); + + Vector2<float> result4 = float_vec1 / float_vec2; + EXPECT_FLOAT_EQ(result4.x, 0.333333343f); + EXPECT_FLOAT_EQ(result4.y, 0.5f); +} + +TEST_F(Vector2Test, DivideScalar) { + Vector2<int> result = int_vec1 / 2; + EXPECT_EQ(result.x, 0); + EXPECT_EQ(result.y, 1); + + Vector2<double> result2 = double_vec1 / 2.0; + EXPECT_FLOAT_EQ(result2.x, 0.5); + EXPECT_FLOAT_EQ(result2.y, 1.0); + + Vector2<long> result3 = long_vec1 / 2; + EXPECT_EQ(result3.x, 0); + EXPECT_EQ(result3.y, 1); + + Vector2<float> result4 = float_vec1 / 2.0f; + EXPECT_FLOAT_EQ(result4.x, 0.5f); + EXPECT_FLOAT_EQ(result4.y, 1.0f); +} + +TEST_F(Vector2Test, AddChain) { + Vector2<int> result = int_vec1; + result += int_vec2; + EXPECT_EQ(result.x, 4); + EXPECT_EQ(result.y, 6); + + Vector2<double> result2 = double_vec1; + result2 += double_vec2; + EXPECT_FLOAT_EQ(result2.x, 4.0); + EXPECT_FLOAT_EQ(result2.y, 6.0); + + Vector2<long> result3 = long_vec1; + result3 += long_vec2; + EXPECT_EQ(result3.x, 4); + EXPECT_EQ(result3.y, 6); + + Vector2<float> result4 = float_vec1; + result4 += float_vec2; + EXPECT_FLOAT_EQ(result4.x, 4.0f); + EXPECT_FLOAT_EQ(result4.y, 6.0f); +} + +TEST_F(Vector2Test, AddScalarChain) { + Vector2<int> result = int_vec1; + result += 1; + EXPECT_EQ(result.x, 2); + EXPECT_EQ(result.y, 3); + + Vector2<double> result2 = double_vec1; + result2 += 1.0; + EXPECT_FLOAT_EQ(result2.x, 2.0); + EXPECT_FLOAT_EQ(result2.y, 3.0); + + Vector2<long> result3 = long_vec1; + result3 += 1; + EXPECT_EQ(result3.x, 2); + EXPECT_EQ(result3.y, 3); + + Vector2<float> result4 = float_vec1; + result4 += 1.0f; + EXPECT_FLOAT_EQ(result4.x, 2.0f); + EXPECT_FLOAT_EQ(result4.y, 3.0f); +} + +TEST_F(Vector2Test, SubtractChain) { + Vector2<int> result = int_vec1; + result -= int_vec2; + EXPECT_EQ(result.x, -2); + EXPECT_EQ(result.y, -2); + + Vector2<double> result2 = double_vec1; + result2 -= double_vec2; + EXPECT_FLOAT_EQ(result2.x, -2.0); + EXPECT_FLOAT_EQ(result2.y, -2.0); + + Vector2<long> result3 = long_vec1; + result3 -= long_vec2; + EXPECT_EQ(result3.x, -2); + EXPECT_EQ(result3.y, -2); + + Vector2<float> result4 = float_vec1; + result4 -= float_vec2; + EXPECT_FLOAT_EQ(result4.x, -2.0f); + EXPECT_FLOAT_EQ(result4.y, -2.0f); +} + +TEST_F(Vector2Test, SubtractScalarChain) { + Vector2<int> result = int_vec1; + result -= 1; + EXPECT_EQ(result.x, 0); + EXPECT_EQ(result.y, 1); + + Vector2<double> result2 = double_vec1; + result2 -= 1.0; + EXPECT_FLOAT_EQ(result2.x, 0.0); + EXPECT_FLOAT_EQ(result2.y, 1.0); + + Vector2<long> result3 = long_vec1; + result3 -= 1; + EXPECT_EQ(result3.x, 0); + EXPECT_EQ(result3.y, 1); + + Vector2<float> result4 = float_vec1; + result4 -= 1.0f; + EXPECT_FLOAT_EQ(result4.x, 0.0f); + EXPECT_FLOAT_EQ(result4.y, 1.0f); +} + +TEST_F(Vector2Test, MultiplyChain) { + Vector2<int> result = int_vec1; + result *= int_vec2; + EXPECT_EQ(result.x, 3); + EXPECT_EQ(result.y, 8); + + Vector2<double> result2 = double_vec1; + result2 *= double_vec2; + EXPECT_FLOAT_EQ(result2.x, 3.0); + EXPECT_FLOAT_EQ(result2.y, 8.0); + + Vector2<long> result3 = long_vec1; + result3 *= long_vec2; + EXPECT_EQ(result3.x, 3); + EXPECT_EQ(result3.y, 8); + + Vector2<float> result4 = float_vec1; + result4 *= float_vec2; + EXPECT_FLOAT_EQ(result4.x, 3.0f); + EXPECT_FLOAT_EQ(result4.y, 8.0f); +} + +TEST_F(Vector2Test, MultiplyScalarChain) { + Vector2<int> result = int_vec1; + result *= 2; + EXPECT_EQ(result.x, 2); + EXPECT_EQ(result.y, 4); + + Vector2<double> result2 = double_vec1; + result2 *= 2.0; + EXPECT_FLOAT_EQ(result2.x, 2.0); + EXPECT_FLOAT_EQ(result2.y, 4.0); + + Vector2<long> result3 = long_vec1; + result3 *= 2; + EXPECT_EQ(result3.x, 2); + EXPECT_EQ(result3.y, 4); + + Vector2<float> result4 = float_vec1; + result4 *= 2.0f; + EXPECT_FLOAT_EQ(result4.x, 2.0f); + EXPECT_FLOAT_EQ(result4.y, 4.0f); +} + +TEST_F(Vector2Test, DivideChain) { + Vector2<int> result = int_vec1; + result /= int_vec2; + EXPECT_EQ(result.x, 0); + EXPECT_EQ(result.y, 0); + + Vector2<double> result2 = double_vec1; + result2 /= double_vec2; + EXPECT_FLOAT_EQ(result2.x, 0.33333333333333331); + EXPECT_FLOAT_EQ(result2.y, 0.5); + + Vector2<long> result3 = long_vec1; + result3 /= long_vec2; + EXPECT_EQ(result3.x, 0); + EXPECT_EQ(result3.y, 0); + + Vector2<float> result4 = float_vec1; + result4 /= float_vec2; + EXPECT_FLOAT_EQ(result4.x, 0.333333343f); + EXPECT_FLOAT_EQ(result4.y, 0.5f); +} + +TEST_F(Vector2Test, DivideScalarChain) { + Vector2<int> result = int_vec1; + result /= 2; + EXPECT_EQ(result.x, 0); + EXPECT_EQ(result.y, 1); + + Vector2<double> result2 = double_vec1; + result2 /= 2.0; + EXPECT_FLOAT_EQ(result2.x, 0.5); + EXPECT_FLOAT_EQ(result2.y, 1.0); + + Vector2<long> result3 = long_vec1; + result3 /= 2; + EXPECT_EQ(result3.x, 0); + EXPECT_EQ(result3.y, 1); + + Vector2<float> result4 = float_vec1; + result4 /= 2.0f; + EXPECT_FLOAT_EQ(result4.x, 0.5f); + EXPECT_FLOAT_EQ(result4.y, 1.0f); +} + +TEST_F(Vector2Test, Negatation) { + Vector2<int> result = -int_vec1; + EXPECT_EQ(result.x, -1); + EXPECT_EQ(result.y, -2); + + Vector2<double> result2 = -double_vec1; + EXPECT_FLOAT_EQ(result2.x, -1.0); + EXPECT_FLOAT_EQ(result2.y, -2.0); + + Vector2<long> result3 = -long_vec1; + EXPECT_EQ(result3.x, -1); + EXPECT_EQ(result3.y, -2); + + Vector2<float> result4 = -float_vec1; + EXPECT_FLOAT_EQ(result4.x, -1.0f); + EXPECT_FLOAT_EQ(result4.y, -2.0f); +} + +TEST_F(Vector2Test, Equals) { + EXPECT_TRUE(int_vec1 == int_vec1); + EXPECT_FALSE(int_vec1 == int_vec2); + EXPECT_TRUE(double_vec1 == double_vec1); + EXPECT_FALSE(double_vec1 == double_vec2); + EXPECT_TRUE(long_vec1 == long_vec1); + EXPECT_FALSE(long_vec1 == long_vec2); +} + +TEST_F(Vector2Test, NotEquals) { + EXPECT_FALSE(int_vec1 != int_vec1); + EXPECT_TRUE(int_vec1 != int_vec2); + EXPECT_FALSE(double_vec1 != double_vec1); + EXPECT_TRUE(double_vec1 != double_vec2); + EXPECT_FALSE(long_vec1 != long_vec1); + EXPECT_TRUE(long_vec1 != long_vec2); +} |