- First much better solution is to use the implementation of this function
A::getType
- maybe it does not have to be mocked? E.g.: if it just returns some value that is set in constructor - then just construct A
in the way that is needed in your test case:
class A{
public:
A(int a) : a(a) {}
template<typename T>
int getType(int val)
{
return a + val;
}
private:
int a;
};
TEST(...)
{
A a(TYPE_VALUE_FOR_TEST);
...
}
- If it cannot be done that way - then you might consider to have some instrumentation for UT that is switched with preprocessor macros:
#ifdef TESTING
namespace Testing
{
using std::pair<std::type_index, int> AGetTypeKey;
std::map<AGetTypeKey, int> AGetTypeExpectedValues;
template <typename T>
void expectAGetType(int inputValue, int expectedResult)
{
AGetTypeExpectedValues[AGetTypeKey(std::type_index(typeid(T)), inputValue)] = expectedResult;
}
template <typename T>
int getAGetType(int value)
{
return AGetTypeExpectedValues[AGetTypeKey(std::type_index(typeid(T)), inputValue)];
}
}
#endif
class A{
public:
A(int a) : a(a) {}
template<typename T>
int getType(int val)
{
#if TESTING
return Testing::getAGetType<T>(val);
#else
// your "normal" implementation
...
#endif
}
private:
int a;
};
// compiled with -DTESTING=1
#ifndef TESTING
#error ...
#endif
TEST(...)
{
Testing::expectAGetType<float>(EXPECTED_INPUT_VALUE,
TYPE_VALUE_FOR_FLOAT);
...
}
Regarding point-2 - of course all testing code should be carefully separated from "normal code" - e.g. in some separated header files.
It is worth to say that none of these solution is perfect - and this second solution might not be 100% reliable as you would test not real code but some testable version of it.
Maybe you should start from rethinking your design - as it seems the design was not completed with "design for testability" in mind.