mirror of https://github.com/ArduPilot/ardupilot
79 lines
2.3 KiB
C++
79 lines
2.3 KiB
C++
|
#include <AP_gtest.h>
|
||
|
|
||
|
#include <SITL/SIM_MS5525.h>
|
||
|
const AP_HAL::HAL& hal = AP_HAL::get_HAL();
|
||
|
|
||
|
using namespace SITL;
|
||
|
|
||
|
// Dummy class to access protected functions through a public interface for unit tests
|
||
|
class dummy : public SITL::MS5525
|
||
|
{
|
||
|
public:
|
||
|
void convert_forward_pub(int32_t D1, int32_t D2, float &P_Pa, float &Temp_C);
|
||
|
void convert_pub(float P_Pa, float Temp_C, uint32_t &D1, uint32_t &D2);
|
||
|
|
||
|
};
|
||
|
|
||
|
void dummy::convert_forward_pub(int32_t D1, int32_t D2, float &P_Pa, float &Temp_C)
|
||
|
{
|
||
|
dummy::convert_forward(D1, D2, P_Pa, Temp_C);
|
||
|
}
|
||
|
|
||
|
void dummy::convert_pub(float P_Pa, float Temp_C, uint32_t &D1, uint32_t &D2)
|
||
|
{
|
||
|
dummy::convert(P_Pa, Temp_C, D1, D2);
|
||
|
}
|
||
|
|
||
|
const struct forward_check {
|
||
|
uint32_t D1; // pressure bytes
|
||
|
uint32_t D2; // temperature bytes
|
||
|
float P_Pa;
|
||
|
float Temp_C;
|
||
|
} conversion_table [] = {
|
||
|
{9453147, 3438567, 245, -20}, // Negative Temperatures, Positive Pressure
|
||
|
|
||
|
{9407591, 3438567, -245, -20}, // Negative Temperatures, Negative Pressure
|
||
|
|
||
|
{9153883, 3774904, 245, 15.15}, // < 20 deg C Temperatures, Positive Pressure
|
||
|
|
||
|
{9112431, 3774904, -245, 15.15}, // < 20 deg C Temperatures, Negative Pressure
|
||
|
|
||
|
{8979245, 3976802, 1.2250, 36.25}, // > 20 deg C Temperature, Positive Pressure
|
||
|
{8998809, 3976802, 245, 36.25}, // > 20 deg C Temperature, Positive Pressure
|
||
|
{9500811, 3976802, 6500, 36.25}, // > 20 deg C Temperature, Positive Pressure
|
||
|
|
||
|
{8959483, 3976802, -245, 36.25}, // > 20 deg C Temperature, Negative Pressure
|
||
|
};
|
||
|
|
||
|
TEST(MS5525, convert_forward)
|
||
|
{
|
||
|
dummy ms5525;
|
||
|
float accuracy_Pa = 0.05;
|
||
|
float accuracy_Temp_C = 0.01;
|
||
|
|
||
|
for(auto elem: conversion_table) {
|
||
|
float P_Pa;
|
||
|
float Temp_C;
|
||
|
ms5525.convert_forward_pub(elem.D1, elem.D2, P_Pa, Temp_C);
|
||
|
EXPECT_NEAR(P_Pa, elem.P_Pa, accuracy_Pa);
|
||
|
EXPECT_NEAR(Temp_C, elem.Temp_C, accuracy_Temp_C);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
TEST(MS5525, convert)
|
||
|
{
|
||
|
dummy ms5525;
|
||
|
|
||
|
for(auto elem: conversion_table) {
|
||
|
uint32_t D1;
|
||
|
uint32_t D2;
|
||
|
ms5525.convert_pub(elem.P_Pa, elem.Temp_C, D1, D2);
|
||
|
|
||
|
// Expect NEAR here instead of EQ because in 32bit they are off by 1
|
||
|
EXPECT_NEAR(D1, elem.D1, 1);
|
||
|
EXPECT_NEAR(D2, elem.D2, 1);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
AP_GTEST_MAIN()
|