| // Copyright 2007, Google Inc. |
| // All rights reserved. |
| // |
| // Redistribution and use in source and binary forms, with or without |
| // modification, are permitted provided that the following conditions are |
| // met: |
| // |
| // * Redistributions of source code must retain the above copyright |
| // notice, this list of conditions and the following disclaimer. |
| // * Redistributions in binary form must reproduce the above |
| // copyright notice, this list of conditions and the following disclaimer |
| // in the documentation and/or other materials provided with the |
| // distribution. |
| // * Neither the name of Google Inc. nor the names of its |
| // contributors may be used to endorse or promote products derived from |
| // this software without specific prior written permission. |
| // |
| // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| |
| // Google Mock - a framework for writing C++ mock classes. |
| // |
| // This file tests some commonly used argument matchers. |
| |
| // Silence warning C4244: 'initializing': conversion from 'int' to 'short', |
| // possible loss of data and C4100, unreferenced local parameter |
| #ifdef _MSC_VER |
| #pragma warning(push) |
| #pragma warning(disable : 4244) |
| #pragma warning(disable : 4100) |
| #endif |
| |
| #include "test/gmock-matchers_test.h" |
| |
| namespace testing { |
| namespace gmock_matchers_test { |
| namespace { |
| |
| typedef ::std::tuple<long, int> Tuple2; // NOLINT |
| |
| // Tests that Eq() matches a 2-tuple where the first field == the |
| // second field. |
| TEST(Eq2Test, MatchesEqualArguments) { |
| Matcher<const Tuple2&> m = Eq(); |
| EXPECT_TRUE(m.Matches(Tuple2(5L, 5))); |
| EXPECT_FALSE(m.Matches(Tuple2(5L, 6))); |
| } |
| |
| // Tests that Eq() describes itself properly. |
| TEST(Eq2Test, CanDescribeSelf) { |
| Matcher<const Tuple2&> m = Eq(); |
| EXPECT_EQ("are an equal pair", Describe(m)); |
| } |
| |
| // Tests that Ge() matches a 2-tuple where the first field >= the |
| // second field. |
| TEST(Ge2Test, MatchesGreaterThanOrEqualArguments) { |
| Matcher<const Tuple2&> m = Ge(); |
| EXPECT_TRUE(m.Matches(Tuple2(5L, 4))); |
| EXPECT_TRUE(m.Matches(Tuple2(5L, 5))); |
| EXPECT_FALSE(m.Matches(Tuple2(5L, 6))); |
| } |
| |
| // Tests that Ge() describes itself properly. |
| TEST(Ge2Test, CanDescribeSelf) { |
| Matcher<const Tuple2&> m = Ge(); |
| EXPECT_EQ("are a pair where the first >= the second", Describe(m)); |
| } |
| |
| // Tests that Gt() matches a 2-tuple where the first field > the |
| // second field. |
| TEST(Gt2Test, MatchesGreaterThanArguments) { |
| Matcher<const Tuple2&> m = Gt(); |
| EXPECT_TRUE(m.Matches(Tuple2(5L, 4))); |
| EXPECT_FALSE(m.Matches(Tuple2(5L, 5))); |
| EXPECT_FALSE(m.Matches(Tuple2(5L, 6))); |
| } |
| |
| // Tests that Gt() describes itself properly. |
| TEST(Gt2Test, CanDescribeSelf) { |
| Matcher<const Tuple2&> m = Gt(); |
| EXPECT_EQ("are a pair where the first > the second", Describe(m)); |
| } |
| |
| // Tests that Le() matches a 2-tuple where the first field <= the |
| // second field. |
| TEST(Le2Test, MatchesLessThanOrEqualArguments) { |
| Matcher<const Tuple2&> m = Le(); |
| EXPECT_TRUE(m.Matches(Tuple2(5L, 6))); |
| EXPECT_TRUE(m.Matches(Tuple2(5L, 5))); |
| EXPECT_FALSE(m.Matches(Tuple2(5L, 4))); |
| } |
| |
| // Tests that Le() describes itself properly. |
| TEST(Le2Test, CanDescribeSelf) { |
| Matcher<const Tuple2&> m = Le(); |
| EXPECT_EQ("are a pair where the first <= the second", Describe(m)); |
| } |
| |
| // Tests that Lt() matches a 2-tuple where the first field < the |
| // second field. |
| TEST(Lt2Test, MatchesLessThanArguments) { |
| Matcher<const Tuple2&> m = Lt(); |
| EXPECT_TRUE(m.Matches(Tuple2(5L, 6))); |
| EXPECT_FALSE(m.Matches(Tuple2(5L, 5))); |
| EXPECT_FALSE(m.Matches(Tuple2(5L, 4))); |
| } |
| |
| // Tests that Lt() describes itself properly. |
| TEST(Lt2Test, CanDescribeSelf) { |
| Matcher<const Tuple2&> m = Lt(); |
| EXPECT_EQ("are a pair where the first < the second", Describe(m)); |
| } |
| |
| // Tests that Ne() matches a 2-tuple where the first field != the |
| // second field. |
| TEST(Ne2Test, MatchesUnequalArguments) { |
| Matcher<const Tuple2&> m = Ne(); |
| EXPECT_TRUE(m.Matches(Tuple2(5L, 6))); |
| EXPECT_TRUE(m.Matches(Tuple2(5L, 4))); |
| EXPECT_FALSE(m.Matches(Tuple2(5L, 5))); |
| } |
| |
| // Tests that Ne() describes itself properly. |
| TEST(Ne2Test, CanDescribeSelf) { |
| Matcher<const Tuple2&> m = Ne(); |
| EXPECT_EQ("are an unequal pair", Describe(m)); |
| } |
| |
| TEST(PairMatchBaseTest, WorksWithMoveOnly) { |
| using Pointers = std::tuple<std::unique_ptr<int>, std::unique_ptr<int>>; |
| Matcher<Pointers> matcher = Eq(); |
| Pointers pointers; |
| // Tested values don't matter; the point is that matcher does not copy the |
| // matched values. |
| EXPECT_TRUE(matcher.Matches(pointers)); |
| } |
| |
| // Tests that IsNan() matches a NaN, with float. |
| TEST(IsNan, FloatMatchesNan) { |
| float quiet_nan = std::numeric_limits<float>::quiet_NaN(); |
| float other_nan = std::nanf("1"); |
| float real_value = 1.0f; |
| |
| Matcher<float> m = IsNan(); |
| EXPECT_TRUE(m.Matches(quiet_nan)); |
| EXPECT_TRUE(m.Matches(other_nan)); |
| EXPECT_FALSE(m.Matches(real_value)); |
| |
| Matcher<float&> m_ref = IsNan(); |
| EXPECT_TRUE(m_ref.Matches(quiet_nan)); |
| EXPECT_TRUE(m_ref.Matches(other_nan)); |
| EXPECT_FALSE(m_ref.Matches(real_value)); |
| |
| Matcher<const float&> m_cref = IsNan(); |
| EXPECT_TRUE(m_cref.Matches(quiet_nan)); |
| EXPECT_TRUE(m_cref.Matches(other_nan)); |
| EXPECT_FALSE(m_cref.Matches(real_value)); |
| } |
| |
| // Tests that IsNan() matches a NaN, with double. |
| TEST(IsNan, DoubleMatchesNan) { |
| double quiet_nan = std::numeric_limits<double>::quiet_NaN(); |
| double other_nan = std::nan("1"); |
| double real_value = 1.0; |
| |
| Matcher<double> m = IsNan(); |
| EXPECT_TRUE(m.Matches(quiet_nan)); |
| EXPECT_TRUE(m.Matches(other_nan)); |
| EXPECT_FALSE(m.Matches(real_value)); |
| |
| Matcher<double&> m_ref = IsNan(); |
| EXPECT_TRUE(m_ref.Matches(quiet_nan)); |
| EXPECT_TRUE(m_ref.Matches(other_nan)); |
| EXPECT_FALSE(m_ref.Matches(real_value)); |
| |
| Matcher<const double&> m_cref = IsNan(); |
| EXPECT_TRUE(m_cref.Matches(quiet_nan)); |
| EXPECT_TRUE(m_cref.Matches(other_nan)); |
| EXPECT_FALSE(m_cref.Matches(real_value)); |
| } |
| |
| // Tests that IsNan() matches a NaN, with long double. |
| TEST(IsNan, LongDoubleMatchesNan) { |
| long double quiet_nan = std::numeric_limits<long double>::quiet_NaN(); |
| long double other_nan = std::nan("1"); |
| long double real_value = 1.0; |
| |
| Matcher<long double> m = IsNan(); |
| EXPECT_TRUE(m.Matches(quiet_nan)); |
| EXPECT_TRUE(m.Matches(other_nan)); |
| EXPECT_FALSE(m.Matches(real_value)); |
| |
| Matcher<long double&> m_ref = IsNan(); |
| EXPECT_TRUE(m_ref.Matches(quiet_nan)); |
| EXPECT_TRUE(m_ref.Matches(other_nan)); |
| EXPECT_FALSE(m_ref.Matches(real_value)); |
| |
| Matcher<const long double&> m_cref = IsNan(); |
| EXPECT_TRUE(m_cref.Matches(quiet_nan)); |
| EXPECT_TRUE(m_cref.Matches(other_nan)); |
| EXPECT_FALSE(m_cref.Matches(real_value)); |
| } |
| |
| // Tests that IsNan() works with Not. |
| TEST(IsNan, NotMatchesNan) { |
| Matcher<float> mf = Not(IsNan()); |
| EXPECT_FALSE(mf.Matches(std::numeric_limits<float>::quiet_NaN())); |
| EXPECT_FALSE(mf.Matches(std::nanf("1"))); |
| EXPECT_TRUE(mf.Matches(1.0)); |
| |
| Matcher<double> md = Not(IsNan()); |
| EXPECT_FALSE(md.Matches(std::numeric_limits<double>::quiet_NaN())); |
| EXPECT_FALSE(md.Matches(std::nan("1"))); |
| EXPECT_TRUE(md.Matches(1.0)); |
| |
| Matcher<long double> mld = Not(IsNan()); |
| EXPECT_FALSE(mld.Matches(std::numeric_limits<long double>::quiet_NaN())); |
| EXPECT_FALSE(mld.Matches(std::nanl("1"))); |
| EXPECT_TRUE(mld.Matches(1.0)); |
| } |
| |
| // Tests that IsNan() can describe itself. |
| TEST(IsNan, CanDescribeSelf) { |
| Matcher<float> mf = IsNan(); |
| EXPECT_EQ("is NaN", Describe(mf)); |
| |
| Matcher<double> md = IsNan(); |
| EXPECT_EQ("is NaN", Describe(md)); |
| |
| Matcher<long double> mld = IsNan(); |
| EXPECT_EQ("is NaN", Describe(mld)); |
| } |
| |
| // Tests that IsNan() can describe itself with Not. |
| TEST(IsNan, CanDescribeSelfWithNot) { |
| Matcher<float> mf = Not(IsNan()); |
| EXPECT_EQ("isn't NaN", Describe(mf)); |
| |
| Matcher<double> md = Not(IsNan()); |
| EXPECT_EQ("isn't NaN", Describe(md)); |
| |
| Matcher<long double> mld = Not(IsNan()); |
| EXPECT_EQ("isn't NaN", Describe(mld)); |
| } |
| |
| // Tests that FloatEq() matches a 2-tuple where |
| // FloatEq(first field) matches the second field. |
| TEST(FloatEq2Test, MatchesEqualArguments) { |
| typedef ::std::tuple<float, float> Tpl; |
| Matcher<const Tpl&> m = FloatEq(); |
| EXPECT_TRUE(m.Matches(Tpl(1.0f, 1.0f))); |
| EXPECT_TRUE(m.Matches(Tpl(0.3f, 0.1f + 0.1f + 0.1f))); |
| EXPECT_FALSE(m.Matches(Tpl(1.1f, 1.0f))); |
| } |
| |
| // Tests that FloatEq() describes itself properly. |
| TEST(FloatEq2Test, CanDescribeSelf) { |
| Matcher<const ::std::tuple<float, float>&> m = FloatEq(); |
| EXPECT_EQ("are an almost-equal pair", Describe(m)); |
| } |
| |
| // Tests that NanSensitiveFloatEq() matches a 2-tuple where |
| // NanSensitiveFloatEq(first field) matches the second field. |
| TEST(NanSensitiveFloatEqTest, MatchesEqualArgumentsWithNaN) { |
| typedef ::std::tuple<float, float> Tpl; |
| Matcher<const Tpl&> m = NanSensitiveFloatEq(); |
| EXPECT_TRUE(m.Matches(Tpl(1.0f, 1.0f))); |
| EXPECT_TRUE(m.Matches(Tpl(std::numeric_limits<float>::quiet_NaN(), |
| std::numeric_limits<float>::quiet_NaN()))); |
| EXPECT_FALSE(m.Matches(Tpl(1.1f, 1.0f))); |
| EXPECT_FALSE(m.Matches(Tpl(1.0f, std::numeric_limits<float>::quiet_NaN()))); |
| EXPECT_FALSE(m.Matches(Tpl(std::numeric_limits<float>::quiet_NaN(), 1.0f))); |
| } |
| |
| // Tests that NanSensitiveFloatEq() describes itself properly. |
| TEST(NanSensitiveFloatEqTest, CanDescribeSelfWithNaNs) { |
| Matcher<const ::std::tuple<float, float>&> m = NanSensitiveFloatEq(); |
| EXPECT_EQ("are an almost-equal pair", Describe(m)); |
| } |
| |
| // Tests that DoubleEq() matches a 2-tuple where |
| // DoubleEq(first field) matches the second field. |
| TEST(DoubleEq2Test, MatchesEqualArguments) { |
| typedef ::std::tuple<double, double> Tpl; |
| Matcher<const Tpl&> m = DoubleEq(); |
| EXPECT_TRUE(m.Matches(Tpl(1.0, 1.0))); |
| EXPECT_TRUE(m.Matches(Tpl(0.3, 0.1 + 0.1 + 0.1))); |
| EXPECT_FALSE(m.Matches(Tpl(1.1, 1.0))); |
| } |
| |
| // Tests that DoubleEq() describes itself properly. |
| TEST(DoubleEq2Test, CanDescribeSelf) { |
| Matcher<const ::std::tuple<double, double>&> m = DoubleEq(); |
| EXPECT_EQ("are an almost-equal pair", Describe(m)); |
| } |
| |
| // Tests that NanSensitiveDoubleEq() matches a 2-tuple where |
| // NanSensitiveDoubleEq(first field) matches the second field. |
| TEST(NanSensitiveDoubleEqTest, MatchesEqualArgumentsWithNaN) { |
| typedef ::std::tuple<double, double> Tpl; |
| Matcher<const Tpl&> m = NanSensitiveDoubleEq(); |
| EXPECT_TRUE(m.Matches(Tpl(1.0f, 1.0f))); |
| EXPECT_TRUE(m.Matches(Tpl(std::numeric_limits<double>::quiet_NaN(), |
| std::numeric_limits<double>::quiet_NaN()))); |
| EXPECT_FALSE(m.Matches(Tpl(1.1f, 1.0f))); |
| EXPECT_FALSE(m.Matches(Tpl(1.0f, std::numeric_limits<double>::quiet_NaN()))); |
| EXPECT_FALSE(m.Matches(Tpl(std::numeric_limits<double>::quiet_NaN(), 1.0f))); |
| } |
| |
| // Tests that DoubleEq() describes itself properly. |
| TEST(NanSensitiveDoubleEqTest, CanDescribeSelfWithNaNs) { |
| Matcher<const ::std::tuple<double, double>&> m = NanSensitiveDoubleEq(); |
| EXPECT_EQ("are an almost-equal pair", Describe(m)); |
| } |
| |
| // Tests that FloatEq() matches a 2-tuple where |
| // FloatNear(first field, max_abs_error) matches the second field. |
| TEST(FloatNear2Test, MatchesEqualArguments) { |
| typedef ::std::tuple<float, float> Tpl; |
| Matcher<const Tpl&> m = FloatNear(0.5f); |
| EXPECT_TRUE(m.Matches(Tpl(1.0f, 1.0f))); |
| EXPECT_TRUE(m.Matches(Tpl(1.3f, 1.0f))); |
| EXPECT_FALSE(m.Matches(Tpl(1.8f, 1.0f))); |
| } |
| |
| // Tests that FloatNear() describes itself properly. |
| TEST(FloatNear2Test, CanDescribeSelf) { |
| Matcher<const ::std::tuple<float, float>&> m = FloatNear(0.5f); |
| EXPECT_EQ("are an almost-equal pair", Describe(m)); |
| } |
| |
| // Tests that NanSensitiveFloatNear() matches a 2-tuple where |
| // NanSensitiveFloatNear(first field) matches the second field. |
| TEST(NanSensitiveFloatNearTest, MatchesNearbyArgumentsWithNaN) { |
| typedef ::std::tuple<float, float> Tpl; |
| Matcher<const Tpl&> m = NanSensitiveFloatNear(0.5f); |
| EXPECT_TRUE(m.Matches(Tpl(1.0f, 1.0f))); |
| EXPECT_TRUE(m.Matches(Tpl(1.1f, 1.0f))); |
| EXPECT_TRUE(m.Matches(Tpl(std::numeric_limits<float>::quiet_NaN(), |
| std::numeric_limits<float>::quiet_NaN()))); |
| EXPECT_FALSE(m.Matches(Tpl(1.6f, 1.0f))); |
| EXPECT_FALSE(m.Matches(Tpl(1.0f, std::numeric_limits<float>::quiet_NaN()))); |
| EXPECT_FALSE(m.Matches(Tpl(std::numeric_limits<float>::quiet_NaN(), 1.0f))); |
| } |
| |
| // Tests that NanSensitiveFloatNear() describes itself properly. |
| TEST(NanSensitiveFloatNearTest, CanDescribeSelfWithNaNs) { |
| Matcher<const ::std::tuple<float, float>&> m = NanSensitiveFloatNear(0.5f); |
| EXPECT_EQ("are an almost-equal pair", Describe(m)); |
| } |
| |
| // Tests that FloatEq() matches a 2-tuple where |
| // DoubleNear(first field, max_abs_error) matches the second field. |
| TEST(DoubleNear2Test, MatchesEqualArguments) { |
| typedef ::std::tuple<double, double> Tpl; |
| Matcher<const Tpl&> m = DoubleNear(0.5); |
| EXPECT_TRUE(m.Matches(Tpl(1.0, 1.0))); |
| EXPECT_TRUE(m.Matches(Tpl(1.3, 1.0))); |
| EXPECT_FALSE(m.Matches(Tpl(1.8, 1.0))); |
| } |
| |
| // Tests that DoubleNear() describes itself properly. |
| TEST(DoubleNear2Test, CanDescribeSelf) { |
| Matcher<const ::std::tuple<double, double>&> m = DoubleNear(0.5); |
| EXPECT_EQ("are an almost-equal pair", Describe(m)); |
| } |
| |
| // Tests that NanSensitiveDoubleNear() matches a 2-tuple where |
| // NanSensitiveDoubleNear(first field) matches the second field. |
| TEST(NanSensitiveDoubleNearTest, MatchesNearbyArgumentsWithNaN) { |
| typedef ::std::tuple<double, double> Tpl; |
| Matcher<const Tpl&> m = NanSensitiveDoubleNear(0.5f); |
| EXPECT_TRUE(m.Matches(Tpl(1.0f, 1.0f))); |
| EXPECT_TRUE(m.Matches(Tpl(1.1f, 1.0f))); |
| EXPECT_TRUE(m.Matches(Tpl(std::numeric_limits<double>::quiet_NaN(), |
| std::numeric_limits<double>::quiet_NaN()))); |
| EXPECT_FALSE(m.Matches(Tpl(1.6f, 1.0f))); |
| EXPECT_FALSE(m.Matches(Tpl(1.0f, std::numeric_limits<double>::quiet_NaN()))); |
| EXPECT_FALSE(m.Matches(Tpl(std::numeric_limits<double>::quiet_NaN(), 1.0f))); |
| } |
| |
| // Tests that NanSensitiveDoubleNear() describes itself properly. |
| TEST(NanSensitiveDoubleNearTest, CanDescribeSelfWithNaNs) { |
| Matcher<const ::std::tuple<double, double>&> m = NanSensitiveDoubleNear(0.5f); |
| EXPECT_EQ("are an almost-equal pair", Describe(m)); |
| } |
| |
| // Tests that Not(m) matches any value that doesn't match m. |
| TEST(NotTest, NegatesMatcher) { |
| Matcher<int> m; |
| m = Not(Eq(2)); |
| EXPECT_TRUE(m.Matches(3)); |
| EXPECT_FALSE(m.Matches(2)); |
| } |
| |
| // Tests that Not(m) describes itself properly. |
| TEST(NotTest, CanDescribeSelf) { |
| Matcher<int> m = Not(Eq(5)); |
| EXPECT_EQ("isn't equal to 5", Describe(m)); |
| } |
| |
| // Tests that monomorphic matchers are safely cast by the Not matcher. |
| TEST(NotTest, NotMatcherSafelyCastsMonomorphicMatchers) { |
| // greater_than_5 is a monomorphic matcher. |
| Matcher<int> greater_than_5 = Gt(5); |
| |
| Matcher<const int&> m = Not(greater_than_5); |
| Matcher<int&> m2 = Not(greater_than_5); |
| Matcher<int&> m3 = Not(m); |
| } |
| |
| // Helper to allow easy testing of AllOf matchers with num parameters. |
| void AllOfMatches(int num, const Matcher<int>& m) { |
| SCOPED_TRACE(Describe(m)); |
| EXPECT_TRUE(m.Matches(0)); |
| for (int i = 1; i <= num; ++i) { |
| EXPECT_FALSE(m.Matches(i)); |
| } |
| EXPECT_TRUE(m.Matches(num + 1)); |
| } |
| |
| INSTANTIATE_GTEST_MATCHER_TEST_P(AllOfTest); |
| |
| // Tests that AllOf(m1, ..., mn) matches any value that matches all of |
| // the given matchers. |
| TEST(AllOfTest, MatchesWhenAllMatch) { |
| Matcher<int> m; |
| m = AllOf(Le(2), Ge(1)); |
| EXPECT_TRUE(m.Matches(1)); |
| EXPECT_TRUE(m.Matches(2)); |
| EXPECT_FALSE(m.Matches(0)); |
| EXPECT_FALSE(m.Matches(3)); |
| |
| m = AllOf(Gt(0), Ne(1), Ne(2)); |
| EXPECT_TRUE(m.Matches(3)); |
| EXPECT_FALSE(m.Matches(2)); |
| EXPECT_FALSE(m.Matches(1)); |
| EXPECT_FALSE(m.Matches(0)); |
| |
| m = AllOf(Gt(0), Ne(1), Ne(2), Ne(3)); |
| EXPECT_TRUE(m.Matches(4)); |
| EXPECT_FALSE(m.Matches(3)); |
| EXPECT_FALSE(m.Matches(2)); |
| EXPECT_FALSE(m.Matches(1)); |
| EXPECT_FALSE(m.Matches(0)); |
| |
| m = AllOf(Ge(0), Lt(10), Ne(3), Ne(5), Ne(7)); |
| EXPECT_TRUE(m.Matches(0)); |
| EXPECT_TRUE(m.Matches(1)); |
| EXPECT_FALSE(m.Matches(3)); |
| |
| // The following tests for varying number of sub-matchers. Due to the way |
| // the sub-matchers are handled it is enough to test every sub-matcher once |
| // with sub-matchers using the same matcher type. Varying matcher types are |
| // checked for above. |
| AllOfMatches(2, AllOf(Ne(1), Ne(2))); |
| AllOfMatches(3, AllOf(Ne(1), Ne(2), Ne(3))); |
| AllOfMatches(4, AllOf(Ne(1), Ne(2), Ne(3), Ne(4))); |
| AllOfMatches(5, AllOf(Ne(1), Ne(2), Ne(3), Ne(4), Ne(5))); |
| AllOfMatches(6, AllOf(Ne(1), Ne(2), Ne(3), Ne(4), Ne(5), Ne(6))); |
| AllOfMatches(7, AllOf(Ne(1), Ne(2), Ne(3), Ne(4), Ne(5), Ne(6), Ne(7))); |
| AllOfMatches(8, |
| AllOf(Ne(1), Ne(2), Ne(3), Ne(4), Ne(5), Ne(6), Ne(7), Ne(8))); |
| AllOfMatches( |
| 9, AllOf(Ne(1), Ne(2), Ne(3), Ne(4), Ne(5), Ne(6), Ne(7), Ne(8), Ne(9))); |
| AllOfMatches(10, AllOf(Ne(1), Ne(2), Ne(3), Ne(4), Ne(5), Ne(6), Ne(7), Ne(8), |
| Ne(9), Ne(10))); |
| AllOfMatches( |
| 50, AllOf(Ne(1), Ne(2), Ne(3), Ne(4), Ne(5), Ne(6), Ne(7), Ne(8), Ne(9), |
| Ne(10), Ne(11), Ne(12), Ne(13), Ne(14), Ne(15), Ne(16), Ne(17), |
| Ne(18), Ne(19), Ne(20), Ne(21), Ne(22), Ne(23), Ne(24), Ne(25), |
| Ne(26), Ne(27), Ne(28), Ne(29), Ne(30), Ne(31), Ne(32), Ne(33), |
| Ne(34), Ne(35), Ne(36), Ne(37), Ne(38), Ne(39), Ne(40), Ne(41), |
| Ne(42), Ne(43), Ne(44), Ne(45), Ne(46), Ne(47), Ne(48), Ne(49), |
| Ne(50))); |
| } |
| |
| // Tests that AllOf(m1, ..., mn) describes itself properly. |
| TEST(AllOfTest, CanDescribeSelf) { |
| Matcher<int> m; |
| m = AllOf(Le(2), Ge(1)); |
| EXPECT_EQ("(is <= 2) and (is >= 1)", Describe(m)); |
| |
| m = AllOf(Gt(0), Ne(1), Ne(2)); |
| std::string expected_descr1 = |
| "(is > 0) and (isn't equal to 1) and (isn't equal to 2)"; |
| EXPECT_EQ(expected_descr1, Describe(m)); |
| |
| m = AllOf(Gt(0), Ne(1), Ne(2), Ne(3)); |
| std::string expected_descr2 = |
| "(is > 0) and (isn't equal to 1) and (isn't equal to 2) and (isn't equal " |
| "to 3)"; |
| EXPECT_EQ(expected_descr2, Describe(m)); |
| |
| m = AllOf(Ge(0), Lt(10), Ne(3), Ne(5), Ne(7)); |
| std::string expected_descr3 = |
| "(is >= 0) and (is < 10) and (isn't equal to 3) and (isn't equal to 5) " |
| "and (isn't equal to 7)"; |
| EXPECT_EQ(expected_descr3, Describe(m)); |
| } |
| |
| // Tests that AllOf(m1, ..., mn) describes its negation properly. |
| TEST(AllOfTest, CanDescribeNegation) { |
| Matcher<int> m; |
| m = AllOf(Le(2), Ge(1)); |
| std::string expected_descr4 = "(isn't <= 2) or (isn't >= 1)"; |
| EXPECT_EQ(expected_descr4, DescribeNegation(m)); |
| |
| m = AllOf(Gt(0), Ne(1), Ne(2)); |
| std::string expected_descr5 = |
| "(isn't > 0) or (is equal to 1) or (is equal to 2)"; |
| EXPECT_EQ(expected_descr5, DescribeNegation(m)); |
| |
| m = AllOf(Gt(0), Ne(1), Ne(2), Ne(3)); |
| std::string expected_descr6 = |
| "(isn't > 0) or (is equal to 1) or (is equal to 2) or (is equal to 3)"; |
| EXPECT_EQ(expected_descr6, DescribeNegation(m)); |
| |
| m = AllOf(Ge(0), Lt(10), Ne(3), Ne(5), Ne(7)); |
| std::string expected_desr7 = |
| "(isn't >= 0) or (isn't < 10) or (is equal to 3) or (is equal to 5) or " |
| "(is equal to 7)"; |
| EXPECT_EQ(expected_desr7, DescribeNegation(m)); |
| |
| m = AllOf(Ne(1), Ne(2), Ne(3), Ne(4), Ne(5), Ne(6), Ne(7), Ne(8), Ne(9), |
| Ne(10), Ne(11)); |
| AllOf(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11); |
| EXPECT_THAT(Describe(m), EndsWith("and (isn't equal to 11)")); |
| AllOfMatches(11, m); |
| } |
| |
| // Tests that monomorphic matchers are safely cast by the AllOf matcher. |
| TEST(AllOfTest, AllOfMatcherSafelyCastsMonomorphicMatchers) { |
| // greater_than_5 and less_than_10 are monomorphic matchers. |
| Matcher<int> greater_than_5 = Gt(5); |
| Matcher<int> less_than_10 = Lt(10); |
| |
| Matcher<const int&> m = AllOf(greater_than_5, less_than_10); |
| Matcher<int&> m2 = AllOf(greater_than_5, less_than_10); |
| Matcher<int&> m3 = AllOf(greater_than_5, m2); |
| |
| // Tests that BothOf works when composing itself. |
| Matcher<const int&> m4 = AllOf(greater_than_5, less_than_10, less_than_10); |
| Matcher<int&> m5 = AllOf(greater_than_5, less_than_10, less_than_10); |
| } |
| |
| TEST_P(AllOfTestP, ExplainsResult) { |
| Matcher<int> m; |
| |
| // Successful match. Both matchers need to explain. The second |
| // matcher doesn't give an explanation, so only the first matcher's |
| // explanation is printed. |
| m = AllOf(GreaterThan(10), Lt(30)); |
| EXPECT_EQ("which is 15 more than 10", Explain(m, 25)); |
| |
| // Successful match. Both matchers need to explain. |
| m = AllOf(GreaterThan(10), GreaterThan(20)); |
| EXPECT_EQ("which is 20 more than 10, and which is 10 more than 20", |
| Explain(m, 30)); |
| |
| // Successful match. All matchers need to explain. The second |
| // matcher doesn't given an explanation. |
| m = AllOf(GreaterThan(10), Lt(30), GreaterThan(20)); |
| EXPECT_EQ("which is 15 more than 10, and which is 5 more than 20", |
| Explain(m, 25)); |
| |
| // Successful match. All matchers need to explain. |
| m = AllOf(GreaterThan(10), GreaterThan(20), GreaterThan(30)); |
| EXPECT_EQ( |
| "which is 30 more than 10, and which is 20 more than 20, " |
| "and which is 10 more than 30", |
| Explain(m, 40)); |
| |
| // Failed match. The first matcher, which failed, needs to |
| // explain. |
| m = AllOf(GreaterThan(10), GreaterThan(20)); |
| EXPECT_EQ("which is 5 less than 10", Explain(m, 5)); |
| |
| // Failed match. The second matcher, which failed, needs to |
| // explain. Since it doesn't given an explanation, nothing is |
| // printed. |
| m = AllOf(GreaterThan(10), Lt(30)); |
| EXPECT_EQ("", Explain(m, 40)); |
| |
| // Failed match. The second matcher, which failed, needs to |
| // explain. |
| m = AllOf(GreaterThan(10), GreaterThan(20)); |
| EXPECT_EQ("which is 5 less than 20", Explain(m, 15)); |
| } |
| |
| // Helper to allow easy testing of AnyOf matchers with num parameters. |
| static void AnyOfMatches(int num, const Matcher<int>& m) { |
| SCOPED_TRACE(Describe(m)); |
| EXPECT_FALSE(m.Matches(0)); |
| for (int i = 1; i <= num; ++i) { |
| EXPECT_TRUE(m.Matches(i)); |
| } |
| EXPECT_FALSE(m.Matches(num + 1)); |
| } |
| |
| static void AnyOfStringMatches(int num, const Matcher<std::string>& m) { |
| SCOPED_TRACE(Describe(m)); |
| EXPECT_FALSE(m.Matches(std::to_string(0))); |
| |
| for (int i = 1; i <= num; ++i) { |
| EXPECT_TRUE(m.Matches(std::to_string(i))); |
| } |
| EXPECT_FALSE(m.Matches(std::to_string(num + 1))); |
| } |
| |
| INSTANTIATE_GTEST_MATCHER_TEST_P(AnyOfTest); |
| |
| // Tests that AnyOf(m1, ..., mn) matches any value that matches at |
| // least one of the given matchers. |
| TEST(AnyOfTest, MatchesWhenAnyMatches) { |
| Matcher<int> m; |
| m = AnyOf(Le(1), Ge(3)); |
| EXPECT_TRUE(m.Matches(1)); |
| EXPECT_TRUE(m.Matches(4)); |
| EXPECT_FALSE(m.Matches(2)); |
| |
| m = AnyOf(Lt(0), Eq(1), Eq(2)); |
| EXPECT_TRUE(m.Matches(-1)); |
| EXPECT_TRUE(m.Matches(1)); |
| EXPECT_TRUE(m.Matches(2)); |
| EXPECT_FALSE(m.Matches(0)); |
| |
| m = AnyOf(Lt(0), Eq(1), Eq(2), Eq(3)); |
| EXPECT_TRUE(m.Matches(-1)); |
| EXPECT_TRUE(m.Matches(1)); |
| EXPECT_TRUE(m.Matches(2)); |
| EXPECT_TRUE(m.Matches(3)); |
| EXPECT_FALSE(m.Matches(0)); |
| |
| m = AnyOf(Le(0), Gt(10), 3, 5, 7); |
| EXPECT_TRUE(m.Matches(0)); |
| EXPECT_TRUE(m.Matches(11)); |
| EXPECT_TRUE(m.Matches(3)); |
| EXPECT_FALSE(m.Matches(2)); |
| |
| // The following tests for varying number of sub-matchers. Due to the way |
| // the sub-matchers are handled it is enough to test every sub-matcher once |
| // with sub-matchers using the same matcher type. Varying matcher types are |
| // checked for above. |
| AnyOfMatches(2, AnyOf(1, 2)); |
| AnyOfMatches(3, AnyOf(1, 2, 3)); |
| AnyOfMatches(4, AnyOf(1, 2, 3, 4)); |
| AnyOfMatches(5, AnyOf(1, 2, 3, 4, 5)); |
| AnyOfMatches(6, AnyOf(1, 2, 3, 4, 5, 6)); |
| AnyOfMatches(7, AnyOf(1, 2, 3, 4, 5, 6, 7)); |
| AnyOfMatches(8, AnyOf(1, 2, 3, 4, 5, 6, 7, 8)); |
| AnyOfMatches(9, AnyOf(1, 2, 3, 4, 5, 6, 7, 8, 9)); |
| AnyOfMatches(10, AnyOf(1, 2, 3, 4, 5, 6, 7, 8, 9, 10)); |
| } |
| |
| // Tests the variadic version of the AnyOfMatcher. |
| TEST(AnyOfTest, VariadicMatchesWhenAnyMatches) { |
| // Also make sure AnyOf is defined in the right namespace and does not depend |
| // on ADL. |
| Matcher<int> m = ::testing::AnyOf(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11); |
| |
| EXPECT_THAT(Describe(m), EndsWith("or (is equal to 11)")); |
| AnyOfMatches(11, m); |
| AnyOfMatches(50, AnyOf(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, |
| 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, |
| 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, |
| 45, 46, 47, 48, 49, 50)); |
| AnyOfStringMatches( |
| 50, AnyOf("1", "2", "3", "4", "5", "6", "7", "8", "9", "10", "11", "12", |
| "13", "14", "15", "16", "17", "18", "19", "20", "21", "22", |
| "23", "24", "25", "26", "27", "28", "29", "30", "31", "32", |
| "33", "34", "35", "36", "37", "38", "39", "40", "41", "42", |
| "43", "44", "45", "46", "47", "48", "49", "50")); |
| } |
| |
| TEST(ConditionalTest, MatchesFirstIfCondition) { |
| Matcher<std::string> eq_red = Eq("red"); |
| Matcher<std::string> ne_red = Ne("red"); |
| Matcher<std::string> m = Conditional(true, eq_red, ne_red); |
| EXPECT_TRUE(m.Matches("red")); |
| EXPECT_FALSE(m.Matches("green")); |
| |
| StringMatchResultListener listener; |
| StringMatchResultListener expected; |
| EXPECT_FALSE(m.MatchAndExplain("green", &listener)); |
| EXPECT_FALSE(eq_red.MatchAndExplain("green", &expected)); |
| EXPECT_THAT(listener.str(), Eq(expected.str())); |
| } |
| |
| TEST(ConditionalTest, MatchesSecondIfCondition) { |
| Matcher<std::string> eq_red = Eq("red"); |
| Matcher<std::string> ne_red = Ne("red"); |
| Matcher<std::string> m = Conditional(false, eq_red, ne_red); |
| EXPECT_FALSE(m.Matches("red")); |
| EXPECT_TRUE(m.Matches("green")); |
| |
| StringMatchResultListener listener; |
| StringMatchResultListener expected; |
| EXPECT_FALSE(m.MatchAndExplain("red", &listener)); |
| EXPECT_FALSE(ne_red.MatchAndExplain("red", &expected)); |
| EXPECT_THAT(listener.str(), Eq(expected.str())); |
| } |
| |
| // Tests that AnyOf(m1, ..., mn) describes itself properly. |
| TEST(AnyOfTest, CanDescribeSelf) { |
| Matcher<int> m; |
| m = AnyOf(Le(1), Ge(3)); |
| |
| EXPECT_EQ("(is <= 1) or (is >= 3)", Describe(m)); |
| |
| m = AnyOf(Lt(0), Eq(1), Eq(2)); |
| EXPECT_EQ("(is < 0) or (is equal to 1) or (is equal to 2)", Describe(m)); |
| |
| m = AnyOf(Lt(0), Eq(1), Eq(2), Eq(3)); |
| EXPECT_EQ("(is < 0) or (is equal to 1) or (is equal to 2) or (is equal to 3)", |
| Describe(m)); |
| |
| m = AnyOf(Le(0), Gt(10), 3, 5, 7); |
| EXPECT_EQ( |
| "(is <= 0) or (is > 10) or (is equal to 3) or (is equal to 5) or (is " |
| "equal to 7)", |
| Describe(m)); |
| } |
| |
| // Tests that AnyOf(m1, ..., mn) describes its negation properly. |
| TEST(AnyOfTest, CanDescribeNegation) { |
| Matcher<int> m; |
| m = AnyOf(Le(1), Ge(3)); |
| EXPECT_EQ("(isn't <= 1) and (isn't >= 3)", DescribeNegation(m)); |
| |
| m = AnyOf(Lt(0), Eq(1), Eq(2)); |
| EXPECT_EQ("(isn't < 0) and (isn't equal to 1) and (isn't equal to 2)", |
| DescribeNegation(m)); |
| |
| m = AnyOf(Lt(0), Eq(1), Eq(2), Eq(3)); |
| EXPECT_EQ( |
| "(isn't < 0) and (isn't equal to 1) and (isn't equal to 2) and (isn't " |
| "equal to 3)", |
| DescribeNegation(m)); |
| |
| m = AnyOf(Le(0), Gt(10), 3, 5, 7); |
| EXPECT_EQ( |
| "(isn't <= 0) and (isn't > 10) and (isn't equal to 3) and (isn't equal " |
| "to 5) and (isn't equal to 7)", |
| DescribeNegation(m)); |
| } |
| |
| // Tests that monomorphic matchers are safely cast by the AnyOf matcher. |
| TEST(AnyOfTest, AnyOfMatcherSafelyCastsMonomorphicMatchers) { |
| // greater_than_5 and less_than_10 are monomorphic matchers. |
| Matcher<int> greater_than_5 = Gt(5); |
| Matcher<int> less_than_10 = Lt(10); |
| |
| Matcher<const int&> m = AnyOf(greater_than_5, less_than_10); |
| Matcher<int&> m2 = AnyOf(greater_than_5, less_than_10); |
| Matcher<int&> m3 = AnyOf(greater_than_5, m2); |
| |
| // Tests that EitherOf works when composing itself. |
| Matcher<const int&> m4 = AnyOf(greater_than_5, less_than_10, less_than_10); |
| Matcher<int&> m5 = AnyOf(greater_than_5, less_than_10, less_than_10); |
| } |
| |
| TEST_P(AnyOfTestP, ExplainsResult) { |
| Matcher<int> m; |
| |
| // Failed match. Both matchers need to explain. The second |
| // matcher doesn't give an explanation, so only the first matcher's |
| // explanation is printed. |
| m = AnyOf(GreaterThan(10), Lt(0)); |
| EXPECT_EQ("which is 5 less than 10", Explain(m, 5)); |
| |
| // Failed match. Both matchers need to explain. |
| m = AnyOf(GreaterThan(10), GreaterThan(20)); |
| EXPECT_EQ("which is 5 less than 10, and which is 15 less than 20", |
| Explain(m, 5)); |
| |
| // Failed match. All matchers need to explain. The second |
| // matcher doesn't given an explanation. |
| m = AnyOf(GreaterThan(10), Gt(20), GreaterThan(30)); |
| EXPECT_EQ("which is 5 less than 10, and which is 25 less than 30", |
| Explain(m, 5)); |
| |
| // Failed match. All matchers need to explain. |
| m = AnyOf(GreaterThan(10), GreaterThan(20), GreaterThan(30)); |
| EXPECT_EQ( |
| "which is 5 less than 10, and which is 15 less than 20, " |
| "and which is 25 less than 30", |
| Explain(m, 5)); |
| |
| // Successful match. The first matcher, which succeeded, needs to |
| // explain. |
| m = AnyOf(GreaterThan(10), GreaterThan(20)); |
| EXPECT_EQ("which is 5 more than 10", Explain(m, 15)); |
| |
| // Successful match. The second matcher, which succeeded, needs to |
| // explain. Since it doesn't given an explanation, nothing is |
| // printed. |
| m = AnyOf(GreaterThan(10), Lt(30)); |
| EXPECT_EQ("", Explain(m, 0)); |
| |
| // Successful match. The second matcher, which succeeded, needs to |
| // explain. |
| m = AnyOf(GreaterThan(30), GreaterThan(20)); |
| EXPECT_EQ("which is 5 more than 20", Explain(m, 25)); |
| } |
| |
| // The following predicate function and predicate functor are for |
| // testing the Truly(predicate) matcher. |
| |
| // Returns non-zero if the input is positive. Note that the return |
| // type of this function is not bool. It's OK as Truly() accepts any |
| // unary function or functor whose return type can be implicitly |
| // converted to bool. |
| int IsPositive(double x) { return x > 0 ? 1 : 0; } |
| |
| // This functor returns true if the input is greater than the given |
| // number. |
| class IsGreaterThan { |
| public: |
| explicit IsGreaterThan(int threshold) : threshold_(threshold) {} |
| |
| bool operator()(int n) const { return n > threshold_; } |
| |
| private: |
| int threshold_; |
| }; |
| |
| // For testing Truly(). |
| const int foo = 0; |
| |
| // This predicate returns true if and only if the argument references foo and |
| // has a zero value. |
| bool ReferencesFooAndIsZero(const int& n) { return (&n == &foo) && (n == 0); } |
| |
| // Tests that Truly(predicate) matches what satisfies the given |
| // predicate. |
| TEST(TrulyTest, MatchesWhatSatisfiesThePredicate) { |
| Matcher<double> m = Truly(IsPositive); |
| EXPECT_TRUE(m.Matches(2.0)); |
| EXPECT_FALSE(m.Matches(-1.5)); |
| } |
| |
| // Tests that Truly(predicate_functor) works too. |
| TEST(TrulyTest, CanBeUsedWithFunctor) { |
| Matcher<int> m = Truly(IsGreaterThan(5)); |
| EXPECT_TRUE(m.Matches(6)); |
| EXPECT_FALSE(m.Matches(4)); |
| } |
| |
| // A class that can be implicitly converted to bool. |
| class ConvertibleToBool { |
| public: |
| explicit ConvertibleToBool(int number) : number_(number) {} |
| operator bool() const { return number_ != 0; } |
| |
| private: |
| int number_; |
| }; |
| |
| ConvertibleToBool IsNotZero(int number) { return ConvertibleToBool(number); } |
| |
| // Tests that the predicate used in Truly() may return a class that's |
| // implicitly convertible to bool, even when the class has no |
| // operator!(). |
| TEST(TrulyTest, PredicateCanReturnAClassConvertibleToBool) { |
| Matcher<int> m = Truly(IsNotZero); |
| EXPECT_TRUE(m.Matches(1)); |
| EXPECT_FALSE(m.Matches(0)); |
| } |
| |
| // Tests that Truly(predicate) can describe itself properly. |
| TEST(TrulyTest, CanDescribeSelf) { |
| Matcher<double> m = Truly(IsPositive); |
| EXPECT_EQ("satisfies the given predicate", Describe(m)); |
| } |
| |
| // Tests that Truly(predicate) works when the matcher takes its |
| // argument by reference. |
| TEST(TrulyTest, WorksForByRefArguments) { |
| Matcher<const int&> m = Truly(ReferencesFooAndIsZero); |
| EXPECT_TRUE(m.Matches(foo)); |
| int n = 0; |
| EXPECT_FALSE(m.Matches(n)); |
| } |
| |
| // Tests that Truly(predicate) provides a helpful reason when it fails. |
| TEST(TrulyTest, ExplainsFailures) { |
| StringMatchResultListener listener; |
| EXPECT_FALSE(ExplainMatchResult(Truly(IsPositive), -1, &listener)); |
| EXPECT_EQ(listener.str(), "didn't satisfy the given predicate"); |
| } |
| |
| // Tests that Matches(m) is a predicate satisfied by whatever that |
| // matches matcher m. |
| TEST(MatchesTest, IsSatisfiedByWhatMatchesTheMatcher) { |
| EXPECT_TRUE(Matches(Ge(0))(1)); |
| EXPECT_FALSE(Matches(Eq('a'))('b')); |
| } |
| |
| // Tests that Matches(m) works when the matcher takes its argument by |
| // reference. |
| TEST(MatchesTest, WorksOnByRefArguments) { |
| int m = 0, n = 0; |
| EXPECT_TRUE(Matches(AllOf(Ref(n), Eq(0)))(n)); |
| EXPECT_FALSE(Matches(Ref(m))(n)); |
| } |
| |
| // Tests that a Matcher on non-reference type can be used in |
| // Matches(). |
| TEST(MatchesTest, WorksWithMatcherOnNonRefType) { |
| Matcher<int> eq5 = Eq(5); |
| EXPECT_TRUE(Matches(eq5)(5)); |
| EXPECT_FALSE(Matches(eq5)(2)); |
| } |
| |
| // Tests Value(value, matcher). Since Value() is a simple wrapper for |
| // Matches(), which has been tested already, we don't spend a lot of |
| // effort on testing Value(). |
| TEST(ValueTest, WorksWithPolymorphicMatcher) { |
| EXPECT_TRUE(Value("hi", StartsWith("h"))); |
| EXPECT_FALSE(Value(5, Gt(10))); |
| } |
| |
| TEST(ValueTest, WorksWithMonomorphicMatcher) { |
| const Matcher<int> is_zero = Eq(0); |
| EXPECT_TRUE(Value(0, is_zero)); |
| EXPECT_FALSE(Value('a', is_zero)); |
| |
| int n = 0; |
| const Matcher<const int&> ref_n = Ref(n); |
| EXPECT_TRUE(Value(n, ref_n)); |
| EXPECT_FALSE(Value(1, ref_n)); |
| } |
| |
| TEST(AllArgsTest, WorksForTuple) { |
| EXPECT_THAT(std::make_tuple(1, 2L), AllArgs(Lt())); |
| EXPECT_THAT(std::make_tuple(2L, 1), Not(AllArgs(Lt()))); |
| } |
| |
| TEST(AllArgsTest, WorksForNonTuple) { |
| EXPECT_THAT(42, AllArgs(Gt(0))); |
| EXPECT_THAT('a', Not(AllArgs(Eq('b')))); |
| } |
| |
| class AllArgsHelper { |
| public: |
| AllArgsHelper() {} |
| |
| MOCK_METHOD2(Helper, int(char x, int y)); |
| |
| private: |
| AllArgsHelper(const AllArgsHelper&) = delete; |
| AllArgsHelper& operator=(const AllArgsHelper&) = delete; |
| }; |
| |
| TEST(AllArgsTest, WorksInWithClause) { |
| AllArgsHelper helper; |
| ON_CALL(helper, Helper(_, _)).With(AllArgs(Lt())).WillByDefault(Return(1)); |
| EXPECT_CALL(helper, Helper(_, _)); |
| EXPECT_CALL(helper, Helper(_, _)).With(AllArgs(Gt())).WillOnce(Return(2)); |
| |
| EXPECT_EQ(1, helper.Helper('\1', 2)); |
| EXPECT_EQ(2, helper.Helper('a', 1)); |
| } |
| |
| class OptionalMatchersHelper { |
| public: |
| OptionalMatchersHelper() {} |
| |
| MOCK_METHOD0(NoArgs, int()); |
| |
| MOCK_METHOD1(OneArg, int(int y)); |
| |
| MOCK_METHOD2(TwoArgs, int(char x, int y)); |
| |
| MOCK_METHOD1(Overloaded, int(char x)); |
| MOCK_METHOD2(Overloaded, int(char x, int y)); |
| |
| private: |
| OptionalMatchersHelper(const OptionalMatchersHelper&) = delete; |
| OptionalMatchersHelper& operator=(const OptionalMatchersHelper&) = delete; |
| }; |
| |
| TEST(AllArgsTest, WorksWithoutMatchers) { |
| OptionalMatchersHelper helper; |
| |
| ON_CALL(helper, NoArgs).WillByDefault(Return(10)); |
| ON_CALL(helper, OneArg).WillByDefault(Return(20)); |
| ON_CALL(helper, TwoArgs).WillByDefault(Return(30)); |
| |
| EXPECT_EQ(10, helper.NoArgs()); |
| EXPECT_EQ(20, helper.OneArg(1)); |
| EXPECT_EQ(30, helper.TwoArgs('\1', 2)); |
| |
| EXPECT_CALL(helper, NoArgs).Times(1); |
| EXPECT_CALL(helper, OneArg).WillOnce(Return(100)); |
| EXPECT_CALL(helper, OneArg(17)).WillOnce(Return(200)); |
| EXPECT_CALL(helper, TwoArgs).Times(0); |
| |
| EXPECT_EQ(10, helper.NoArgs()); |
| EXPECT_EQ(100, helper.OneArg(1)); |
| EXPECT_EQ(200, helper.OneArg(17)); |
| } |
| |
| // Tests floating-point matchers. |
| template <typename RawType> |
| class FloatingPointTest : public testing::Test { |
| protected: |
| typedef testing::internal::FloatingPoint<RawType> Floating; |
| typedef typename Floating::Bits Bits; |
| |
| FloatingPointTest() |
| : max_ulps_(Floating::kMaxUlps), |
| zero_bits_(Floating(0).bits()), |
| one_bits_(Floating(1).bits()), |
| infinity_bits_(Floating(Floating::Infinity()).bits()), |
| close_to_positive_zero_( |
| Floating::ReinterpretBits(zero_bits_ + max_ulps_ / 2)), |
| close_to_negative_zero_( |
| -Floating::ReinterpretBits(zero_bits_ + max_ulps_ - max_ulps_ / 2)), |
| further_from_negative_zero_(-Floating::ReinterpretBits( |
| zero_bits_ + max_ulps_ + 1 - max_ulps_ / 2)), |
| close_to_one_(Floating::ReinterpretBits(one_bits_ + max_ulps_)), |
| further_from_one_(Floating::ReinterpretBits(one_bits_ + max_ulps_ + 1)), |
| infinity_(Floating::Infinity()), |
| close_to_infinity_( |
| Floating::ReinterpretBits(infinity_bits_ - max_ulps_)), |
| further_from_infinity_( |
| Floating::ReinterpretBits(infinity_bits_ - max_ulps_ - 1)), |
| max_(Floating::Max()), |
| nan1_(Floating::ReinterpretBits(Floating::kExponentBitMask | 1)), |
| nan2_(Floating::ReinterpretBits(Floating::kExponentBitMask | 200)) {} |
| |
| void TestSize() { EXPECT_EQ(sizeof(RawType), sizeof(Bits)); } |
| |
| // A battery of tests for FloatingEqMatcher::Matches. |
| // matcher_maker is a pointer to a function which creates a FloatingEqMatcher. |
| void TestMatches( |
| testing::internal::FloatingEqMatcher<RawType> (*matcher_maker)(RawType)) { |
| Matcher<RawType> m1 = matcher_maker(0.0); |
| EXPECT_TRUE(m1.Matches(-0.0)); |
| EXPECT_TRUE(m1.Matches(close_to_positive_zero_)); |
| EXPECT_TRUE(m1.Matches(close_to_negative_zero_)); |
| EXPECT_FALSE(m1.Matches(1.0)); |
| |
| Matcher<RawType> m2 = matcher_maker(close_to_positive_zero_); |
| EXPECT_FALSE(m2.Matches(further_from_negative_zero_)); |
| |
| Matcher<RawType> m3 = matcher_maker(1.0); |
| EXPECT_TRUE(m3.Matches(close_to_one_)); |
| EXPECT_FALSE(m3.Matches(further_from_one_)); |
| |
| // Test commutativity: matcher_maker(0.0).Matches(1.0) was tested above. |
| EXPECT_FALSE(m3.Matches(0.0)); |
| |
| Matcher<RawType> m4 = matcher_maker(-infinity_); |
| EXPECT_TRUE(m4.Matches(-close_to_infinity_)); |
| |
| Matcher<RawType> m5 = matcher_maker(infinity_); |
| EXPECT_TRUE(m5.Matches(close_to_infinity_)); |
| |
| // This is interesting as the representations of infinity_ and nan1_ |
| // are only 1 DLP apart. |
| EXPECT_FALSE(m5.Matches(nan1_)); |
| |
| // matcher_maker can produce a Matcher<const RawType&>, which is needed in |
| // some cases. |
| Matcher<const RawType&> m6 = matcher_maker(0.0); |
| EXPECT_TRUE(m6.Matches(-0.0)); |
| EXPECT_TRUE(m6.Matches(close_to_positive_zero_)); |
| EXPECT_FALSE(m6.Matches(1.0)); |
| |
| // matcher_maker can produce a Matcher<RawType&>, which is needed in some |
| // cases. |
| Matcher<RawType&> m7 = matcher_maker(0.0); |
| RawType x = 0.0; |
| EXPECT_TRUE(m7.Matches(x)); |
| x = 0.01f; |
| EXPECT_FALSE(m7.Matches(x)); |
| } |
| |
| // Pre-calculated numbers to be used by the tests. |
| |
| const Bits max_ulps_; |
| |
| const Bits zero_bits_; // The bits that represent 0.0. |
| const Bits one_bits_; // The bits that represent 1.0. |
| const Bits infinity_bits_; // The bits that represent +infinity. |
| |
| // Some numbers close to 0.0. |
| const RawType close_to_positive_zero_; |
| const RawType close_to_negative_zero_; |
| const RawType further_from_negative_zero_; |
| |
| // Some numbers close to 1.0. |
| const RawType close_to_one_; |
| const RawType further_from_one_; |
| |
| // Some numbers close to +infinity. |
| const RawType infinity_; |
| const RawType close_to_infinity_; |
| const RawType further_from_infinity_; |
| |
| // Maximum representable value that's not infinity. |
| const RawType max_; |
| |
| // Some NaNs. |
| const RawType nan1_; |
| const RawType nan2_; |
| }; |
| |
| // Tests floating-point matchers with fixed epsilons. |
| template <typename RawType> |
| class FloatingPointNearTest : public FloatingPointTest<RawType> { |
| protected: |
| typedef FloatingPointTest<RawType> ParentType; |
| |
| // A battery of tests for FloatingEqMatcher::Matches with a fixed epsilon. |
| // matcher_maker is a pointer to a function which creates a FloatingEqMatcher. |
| void TestNearMatches(testing::internal::FloatingEqMatcher<RawType> ( |
| *matcher_maker)(RawType, RawType)) { |
| Matcher<RawType> m1 = matcher_maker(0.0, 0.0); |
| EXPECT_TRUE(m1.Matches(0.0)); |
| EXPECT_TRUE(m1.Matches(-0.0)); |
| EXPECT_FALSE(m1.Matches(ParentType::close_to_positive_zero_)); |
| EXPECT_FALSE(m1.Matches(ParentType::close_to_negative_zero_)); |
| EXPECT_FALSE(m1.Matches(1.0)); |
| |
| Matcher<RawType> m2 = matcher_maker(0.0, 1.0); |
| EXPECT_TRUE(m2.Matches(0.0)); |
| EXPECT_TRUE(m2.Matches(-0.0)); |
| EXPECT_TRUE(m2.Matches(1.0)); |
| EXPECT_TRUE(m2.Matches(-1.0)); |
| EXPECT_FALSE(m2.Matches(ParentType::close_to_one_)); |
| EXPECT_FALSE(m2.Matches(-ParentType::close_to_one_)); |
| |
| // Check that inf matches inf, regardless of the of the specified max |
| // absolute error. |
| Matcher<RawType> m3 = matcher_maker(ParentType::infinity_, 0.0); |
| EXPECT_TRUE(m3.Matches(ParentType::infinity_)); |
| EXPECT_FALSE(m3.Matches(ParentType::close_to_infinity_)); |
| EXPECT_FALSE(m3.Matches(-ParentType::infinity_)); |
| |
| Matcher<RawType> m4 = matcher_maker(-ParentType::infinity_, 0.0); |
| EXPECT_TRUE(m4.Matches(-ParentType::infinity_)); |
| EXPECT_FALSE(m4.Matches(-ParentType::close_to_infinity_)); |
| EXPECT_FALSE(m4.Matches(ParentType::infinity_)); |
| |
| // Test various overflow scenarios. |
| Matcher<RawType> m5 = matcher_maker(ParentType::max_, ParentType::max_); |
| EXPECT_TRUE(m5.Matches(ParentType::max_)); |
| EXPECT_FALSE(m5.Matches(-ParentType::max_)); |
| |
| Matcher<RawType> m6 = matcher_maker(-ParentType::max_, ParentType::max_); |
| EXPECT_FALSE(m6.Matches(ParentType::max_)); |
| EXPECT_TRUE(m6.Matches(-ParentType::max_)); |
| |
| Matcher<RawType> m7 = matcher_maker(ParentType::max_, 0); |
| EXPECT_TRUE(m7.Matches(ParentType::max_)); |
| EXPECT_FALSE(m7.Matches(-ParentType::max_)); |
| |
| Matcher<RawType> m8 = matcher_maker(-ParentType::max_, 0); |
| EXPECT_FALSE(m8.Matches(ParentType::max_)); |
| EXPECT_TRUE(m8.Matches(-ParentType::max_)); |
| |
| // The difference between max() and -max() normally overflows to infinity, |
| // but it should still match if the max_abs_error is also infinity. |
| Matcher<RawType> m9 = |
| matcher_maker(ParentType::max_, ParentType::infinity_); |
| EXPECT_TRUE(m8.Matches(-ParentType::max_)); |
| |
| // matcher_maker can produce a Matcher<const RawType&>, which is needed in |
| // some cases. |
| Matcher<const RawType&> m10 = matcher_maker(0.0, 1.0); |
| EXPECT_TRUE(m10.Matches(-0.0)); |
| EXPECT_TRUE(m10.Matches(ParentType::close_to_positive_zero_)); |
| EXPECT_FALSE(m10.Matches(ParentType::close_to_one_)); |
| |
| // matcher_maker can produce a Matcher<RawType&>, which is needed in some |
| // cases. |
| Matcher<RawType&> m11 = matcher_maker(0.0, 1.0); |
| RawType x = 0.0; |
| EXPECT_TRUE(m11.Matches(x)); |
| x = 1.0f; |
| EXPECT_TRUE(m11.Matches(x)); |
| x = -1.0f; |
| EXPECT_TRUE(m11.Matches(x)); |
| x = 1.1f; |
| EXPECT_FALSE(m11.Matches(x)); |
| x = -1.1f; |
| EXPECT_FALSE(m11.Matches(x)); |
| } |
| }; |
| |
| // Instantiate FloatingPointTest for testing floats. |
| typedef FloatingPointTest<float> FloatTest; |
| |
| TEST_F(FloatTest, FloatEqApproximatelyMatchesFloats) { TestMatches(&FloatEq); } |
| |
| TEST_F(FloatTest, NanSensitiveFloatEqApproximatelyMatchesFloats) { |
| TestMatches(&NanSensitiveFloatEq); |
| } |
| |
| TEST_F(FloatTest, FloatEqCannotMatchNaN) { |
| // FloatEq never matches NaN. |
| Matcher<float> m = FloatEq(nan1_); |
| EXPECT_FALSE(m.Matches(nan1_)); |
| EXPECT_FALSE(m.Matches(nan2_)); |
| EXPECT_FALSE(m.Matches(1.0)); |
| } |
| |
| TEST_F(FloatTest, NanSensitiveFloatEqCanMatchNaN) { |
| // NanSensitiveFloatEq will match NaN. |
| Matcher<float> m = NanSensitiveFloatEq(nan1_); |
| EXPECT_TRUE(m.Matches(nan1_)); |
| EXPECT_TRUE(m.Matches(nan2_)); |
| EXPECT_FALSE(m.Matches(1.0)); |
| } |
| |
| TEST_F(FloatTest, FloatEqCanDescribeSelf) { |
| Matcher<float> m1 = FloatEq(2.0f); |
| EXPECT_EQ("is approximately 2", Describe(m1)); |
| EXPECT_EQ("isn't approximately 2", DescribeNegation(m1)); |
| |
| Matcher<float> m2 = FloatEq(0.5f); |
| EXPECT_EQ("is approximately 0.5", Describe(m2)); |
| EXPECT_EQ("isn't approximately 0.5", DescribeNegation(m2)); |
| |
| Matcher<float> m3 = FloatEq(nan1_); |
| EXPECT_EQ("never matches", Describe(m3)); |
| EXPECT_EQ("is anything", DescribeNegation(m3)); |
| } |
| |
| TEST_F(FloatTest, NanSensitiveFloatEqCanDescribeSelf) { |
| Matcher<float> m1 = NanSensitiveFloatEq(2.0f); |
| EXPECT_EQ("is approximately 2", Describe(m1)); |
| EXPECT_EQ("isn't approximately 2", DescribeNegation(m1)); |
| |
| Matcher<float> m2 = NanSensitiveFloatEq(0.5f); |
| EXPECT_EQ("is approximately 0.5", Describe(m2)); |
| EXPECT_EQ("isn't approximately 0.5", DescribeNegation(m2)); |
| |
| Matcher<float> m3 = NanSensitiveFloatEq(nan1_); |
| EXPECT_EQ("is NaN", Describe(m3)); |
| EXPECT_EQ("isn't NaN", DescribeNegation(m3)); |
| } |
| |
| // Instantiate FloatingPointTest for testing floats with a user-specified |
| // max absolute error. |
| typedef FloatingPointNearTest<float> FloatNearTest; |
| |
| TEST_F(FloatNearTest, FloatNearMatches) { TestNearMatches(&FloatNear); } |
| |
| TEST_F(FloatNearTest, NanSensitiveFloatNearApproximatelyMatchesFloats) { |
| TestNearMatches(&NanSensitiveFloatNear); |
| } |
| |
| TEST_F(FloatNearTest, FloatNearCanDescribeSelf) { |
| Matcher<float> m1 = FloatNear(2.0f, 0.5f); |
| EXPECT_EQ("is approximately 2 (absolute error <= 0.5)", Describe(m1)); |
| EXPECT_EQ("isn't approximately 2 (absolute error > 0.5)", |
| DescribeNegation(m1)); |
| |
| Matcher<float> m2 = FloatNear(0.5f, 0.5f); |
| EXPECT_EQ("is approximately 0.5 (absolute error <= 0.5)", Describe(m2)); |
| EXPECT_EQ("isn't approximately 0.5 (absolute error > 0.5)", |
| DescribeNegation(m2)); |
| |
| Matcher<float> m3 = FloatNear(nan1_, 0.0); |
| EXPECT_EQ("never matches", Describe(m3)); |
| EXPECT_EQ("is anything", DescribeNegation(m3)); |
| } |
| |
| TEST_F(FloatNearTest, NanSensitiveFloatNearCanDescribeSelf) { |
| Matcher<float> m1 = NanSensitiveFloatNear(2.0f, 0.5f); |
| EXPECT_EQ("is approximately 2 (absolute error <= 0.5)", Describe(m1)); |
| EXPECT_EQ("isn't approximately 2 (absolute error > 0.5)", |
| DescribeNegation(m1)); |
| |
| Matcher<float> m2 = NanSensitiveFloatNear(0.5f, 0.5f); |
| EXPECT_EQ("is approximately 0.5 (absolute error <= 0.5)", Describe(m2)); |
| EXPECT_EQ("isn't approximately 0.5 (absolute error > 0.5)", |
| DescribeNegation(m2)); |
| |
| Matcher<float> m3 = NanSensitiveFloatNear(nan1_, 0.1f); |
| EXPECT_EQ("is NaN", Describe(m3)); |
| EXPECT_EQ("isn't NaN", DescribeNegation(m3)); |
| } |
| |
| TEST_F(FloatNearTest, FloatNearCannotMatchNaN) { |
| // FloatNear never matches NaN. |
| Matcher<float> m = FloatNear(ParentType::nan1_, 0.1f); |
| EXPECT_FALSE(m.Matches(nan1_)); |
| EXPECT_FALSE(m.Matches(nan2_)); |
| EXPECT_FALSE(m.Matches(1.0)); |
| } |
| |
| TEST_F(FloatNearTest, NanSensitiveFloatNearCanMatchNaN) { |
| // NanSensitiveFloatNear will match NaN. |
| Matcher<float> m = NanSensitiveFloatNear(nan1_, 0.1f); |
| EXPECT_TRUE(m.Matches(nan1_)); |
| EXPECT_TRUE(m.Matches(nan2_)); |
| EXPECT_FALSE(m.Matches(1.0)); |
| } |
| |
| // Instantiate FloatingPointTest for testing doubles. |
| typedef FloatingPointTest<double> DoubleTest; |
| |
| TEST_F(DoubleTest, DoubleEqApproximatelyMatchesDoubles) { |
| TestMatches(&DoubleEq); |
| } |
| |
| TEST_F(DoubleTest, NanSensitiveDoubleEqApproximatelyMatchesDoubles) { |
| TestMatches(&NanSensitiveDoubleEq); |
| } |
| |
| TEST_F(DoubleTest, DoubleEqCannotMatchNaN) { |
| // DoubleEq never matches NaN. |
| Matcher<double> m = DoubleEq(nan1_); |
| EXPECT_FALSE(m.Matches(nan1_)); |
| EXPECT_FALSE(m.Matches(nan2_)); |
| EXPECT_FALSE(m.Matches(1.0)); |
| } |
| |
| TEST_F(DoubleTest, NanSensitiveDoubleEqCanMatchNaN) { |
| // NanSensitiveDoubleEq will match NaN. |
| Matcher<double> m = NanSensitiveDoubleEq(nan1_); |
| EXPECT_TRUE(m.Matches(nan1_)); |
| EXPECT_TRUE(m.Matches(nan2_)); |
| EXPECT_FALSE(m.Matches(1.0)); |
| } |
| |
| TEST_F(DoubleTest, DoubleEqCanDescribeSelf) { |
| Matcher<double> m1 = DoubleEq(2.0); |
| EXPECT_EQ("is approximately 2", Describe(m1)); |
| EXPECT_EQ("isn't approximately 2", DescribeNegation(m1)); |
| |
| Matcher<double> m2 = DoubleEq(0.5); |
| EXPECT_EQ("is approximately 0.5", Describe(m2)); |
| EXPECT_EQ("isn't approximately 0.5", DescribeNegation(m2)); |
| |
| Matcher<double> m3 = DoubleEq(nan1_); |
| EXPECT_EQ("never matches", Describe(m3)); |
| EXPECT_EQ("is anything", DescribeNegation(m3)); |
| } |
| |
| TEST_F(DoubleTest, NanSensitiveDoubleEqCanDescribeSelf) { |
| Matcher<double> m1 = NanSensitiveDoubleEq(2.0); |
| EXPECT_EQ("is approximately 2", Describe(m1)); |
| EXPECT_EQ("isn't approximately 2", DescribeNegation(m1)); |
| |
| Matcher<double> m2 = NanSensitiveDoubleEq(0.5); |
| EXPECT_EQ("is approximately 0.5", Describe(m2)); |
| EXPECT_EQ("isn't approximately 0.5", DescribeNegation(m2)); |
| |
| Matcher<double> m3 = NanSensitiveDoubleEq(nan1_); |
| EXPECT_EQ("is NaN", Describe(m3)); |
| EXPECT_EQ("isn't NaN", DescribeNegation(m3)); |
| } |
| |
| // Instantiate FloatingPointTest for testing floats with a user-specified |
| // max absolute error. |
| typedef FloatingPointNearTest<double> DoubleNearTest; |
| |
| TEST_F(DoubleNearTest, DoubleNearMatches) { TestNearMatches(&DoubleNear); } |
| |
| TEST_F(DoubleNearTest, NanSensitiveDoubleNearApproximatelyMatchesDoubles) { |
| TestNearMatches(&NanSensitiveDoubleNear); |
| } |
| |
| TEST_F(DoubleNearTest, DoubleNearCanDescribeSelf) { |
| Matcher<double> m1 = DoubleNear(2.0, 0.5); |
| EXPECT_EQ("is approximately 2 (absolute error <= 0.5)", Describe(m1)); |
| EXPECT_EQ("isn't approximately 2 (absolute error > 0.5)", |
| DescribeNegation(m1)); |
| |
| Matcher<double> m2 = DoubleNear(0.5, 0.5); |
| EXPECT_EQ("is approximately 0.5 (absolute error <= 0.5)", Describe(m2)); |
| EXPECT_EQ("isn't approximately 0.5 (absolute error > 0.5)", |
| DescribeNegation(m2)); |
| |
| Matcher<double> m3 = DoubleNear(nan1_, 0.0); |
| EXPECT_EQ("never matches", Describe(m3)); |
| EXPECT_EQ("is anything", DescribeNegation(m3)); |
| } |
| |
| TEST_F(DoubleNearTest, ExplainsResultWhenMatchFails) { |
| EXPECT_EQ("", Explain(DoubleNear(2.0, 0.1), 2.05)); |
| EXPECT_EQ("which is 0.2 from 2", Explain(DoubleNear(2.0, 0.1), 2.2)); |
| EXPECT_EQ("which is -0.3 from 2", Explain(DoubleNear(2.0, 0.1), 1.7)); |
| |
| const std::string explanation = |
| Explain(DoubleNear(2.1, 1e-10), 2.1 + 1.2e-10); |
| // Different C++ implementations may print floating-point numbers |
| // slightly differently. |
| EXPECT_TRUE(explanation == "which is 1.2e-10 from 2.1" || // GCC |
| explanation == "which is 1.2e-010 from 2.1") // MSVC |
| << " where explanation is \"" << explanation << "\"."; |
| } |
| |
| TEST_F(DoubleNearTest, NanSensitiveDoubleNearCanDescribeSelf) { |
| Matcher<double> m1 = NanSensitiveDoubleNear(2.0, 0.5); |
| EXPECT_EQ("is approximately 2 (absolute error <= 0.5)", Describe(m1)); |
| EXPECT_EQ("isn't approximately 2 (absolute error > 0.5)", |
| DescribeNegation(m1)); |
| |
| Matcher<double> m2 = NanSensitiveDoubleNear(0.5, 0.5); |
| EXPECT_EQ("is approximately 0.5 (absolute error <= 0.5)", Describe(m2)); |
| EXPECT_EQ("isn't approximately 0.5 (absolute error > 0.5)", |
| DescribeNegation(m2)); |
| |
| Matcher<double> m3 = NanSensitiveDoubleNear(nan1_, 0.1); |
| EXPECT_EQ("is NaN", Describe(m3)); |
| EXPECT_EQ("isn't NaN", DescribeNegation(m3)); |
| } |
| |
| TEST_F(DoubleNearTest, DoubleNearCannotMatchNaN) { |
| // DoubleNear never matches NaN. |
| Matcher<double> m = DoubleNear(ParentType::nan1_, 0.1); |
| EXPECT_FALSE(m.Matches(nan1_)); |
| EXPECT_FALSE(m.Matches(nan2_)); |
| EXPECT_FALSE(m.Matches(1.0)); |
| } |
| |
| TEST_F(DoubleNearTest, NanSensitiveDoubleNearCanMatchNaN) { |
| // NanSensitiveDoubleNear will match NaN. |
| Matcher<double> m = NanSensitiveDoubleNear(nan1_, 0.1); |
| EXPECT_TRUE(m.Matches(nan1_)); |
| EXPECT_TRUE(m.Matches(nan2_)); |
| EXPECT_FALSE(m.Matches(1.0)); |
| } |
| |
| TEST(NotTest, WorksOnMoveOnlyType) { |
| std::unique_ptr<int> p(new int(3)); |
| EXPECT_THAT(p, Pointee(Eq(3))); |
| EXPECT_THAT(p, Not(Pointee(Eq(2)))); |
| } |
| |
| TEST(AllOfTest, HugeMatcher) { |
| // Verify that using AllOf with many arguments doesn't cause |
| // the compiler to exceed template instantiation depth limit. |
| EXPECT_THAT(0, testing::AllOf(_, _, _, _, _, _, _, _, _, |
| testing::AllOf(_, _, _, _, _, _, _, _, _, _))); |
| } |
| |
| TEST(AnyOfTest, HugeMatcher) { |
| // Verify that using AnyOf with many arguments doesn't cause |
| // the compiler to exceed template instantiation depth limit. |
| EXPECT_THAT(0, testing::AnyOf(_, _, _, _, _, _, _, _, _, |
| testing::AnyOf(_, _, _, _, _, _, _, _, _, _))); |
| } |
| |
| namespace adl_test { |
| |
| // Verifies that the implementation of ::testing::AllOf and ::testing::AnyOf |
| // don't issue unqualified recursive calls. If they do, the argument dependent |
| // name lookup will cause AllOf/AnyOf in the 'adl_test' namespace to be found |
| // as a candidate and the compilation will break due to an ambiguous overload. |
| |
| // The matcher must be in the same namespace as AllOf/AnyOf to make argument |
| // dependent lookup find those. |
| MATCHER(M, "") { |
| (void)arg; |
| return true; |
| } |
| |
| template <typename T1, typename T2> |
| bool AllOf(const T1& /*t1*/, const T2& /*t2*/) { |
| return true; |
| } |
| |
| TEST(AllOfTest, DoesNotCallAllOfUnqualified) { |
| EXPECT_THAT(42, |
| testing::AllOf(M(), M(), M(), M(), M(), M(), M(), M(), M(), M())); |
| } |
| |
| template <typename T1, typename T2> |
| bool AnyOf(const T1&, const T2&) { |
| return true; |
| } |
| |
| TEST(AnyOfTest, DoesNotCallAnyOfUnqualified) { |
| EXPECT_THAT(42, |
| testing::AnyOf(M(), M(), M(), M(), M(), M(), M(), M(), M(), M())); |
| } |
| |
| } // namespace adl_test |
| |
| TEST(AllOfTest, WorksOnMoveOnlyType) { |
| std::unique_ptr<int> p(new int(3)); |
| EXPECT_THAT(p, AllOf(Pointee(Eq(3)), Pointee(Gt(0)), Pointee(Lt(5)))); |
| EXPECT_THAT(p, Not(AllOf(Pointee(Eq(3)), Pointee(Gt(0)), Pointee(Lt(3))))); |
| } |
| |
| TEST(AnyOfTest, WorksOnMoveOnlyType) { |
| std::unique_ptr<int> p(new int(3)); |
| EXPECT_THAT(p, AnyOf(Pointee(Eq(5)), Pointee(Lt(0)), Pointee(Lt(5)))); |
| EXPECT_THAT(p, Not(AnyOf(Pointee(Eq(5)), Pointee(Lt(0)), Pointee(Gt(5))))); |
| } |
| |
| } // namespace |
| } // namespace gmock_matchers_test |
| } // namespace testing |
| |
| #ifdef _MSC_VER |
| #pragma warning(pop) |
| #endif |