#include "config.h" #include "lfptest.h" #include "timevalops.h" #include "unity.h" #include // Required on Solaris for ldexp. void test_Seconds(void) { struct timeval input = {500, 0}; // 500.0 s l_fp expected = {500, 0}; l_fp actual; TVTOTS(&input, &actual); TEST_ASSERT_TRUE(IsEqual(expected, actual)); } void test_MicrosecondsRounded(void) { /* 0.0005 can not be represented exact in a l_fp structure. * It would equal to 2147483,648. This means that * HALF_PROMILLE_UP (which is 2147484) should be * the correct rounding. */ struct timeval input = {0, 500}; // 0.0005 exact l_fp expected = {0, HALF_PROMILLE_UP}; l_fp actual; TVTOTS(&input, &actual); TEST_ASSERT_TRUE(IsEqual(expected, actual)); } void test_MicrosecondsExact(void) { // 0.5 can be represented exact in both l_fp and timeval. const struct timeval input = {10, 500000}; // 0.5 exact const l_fp expected = {10, HALF}; // 0.5 exact l_fp actual; TVTOTS(&input, &actual); // Compare the fractional part with an absolute error given. TEST_ASSERT_EQUAL_UINT(expected.l_ui, actual.l_ui); double expectedDouble, actualDouble; M_LFPTOD(0, expected.l_uf, expectedDouble); M_LFPTOD(0, actual.l_uf, actualDouble); // The error should be less than 0.5 us TEST_ASSERT_DOUBLE_WITHIN(0000005, expectedDouble, actualDouble); }