Assert that code does NOT compile

You're looking for type traits, defined in <type_traits>, to test whether types have certain properties.


A good answer is given at the end of a great article "Diagnosable validity" by Andrzej Krzemieński:

A practical way to check if a given construct fails to compile is to do it from outside C++: prepare a small test program with erroneous construct, compile it, and test if compiler reports compilation failure. This is how “negative” unit tests work with Boost.Build. For an example, see this negative test form Boost.Optional library: optional_test_fail_convert_from_null.cpp. In configuration file it is annotated as compile-fail, meaning that test passes only if compilation fails.


If the goal is to ensure that the code won't compile, you can't have it as part of your test program, since otherwise, your test program won't compile. You have to invoke the compiler on it, and see what the return code is.


template<class T>struct sink{typedef void type;};
template<class T>using sink_t=typename sink<T>::type;

template<typename T, typename=void>struct my_test:std::false_type{};
template<typename T>struct my_test<T,
  sink_t<decltype(

put code here. Note that it must "fail early", ie in the signature of a function, not in the body

  )>
>:std::true_type {};

The above generates a test if the "put code here" can be evaluated.

To determine if "put code here" cannot be evaluated, negate the result of the test.

template<class T>using not_t=std::integral_constant<bool, !T::value>;
not_t< my_test< int > >::value

will be true iff "put code here" fails at the substitution stage. (or you can do it more manually, by swapping std::true_type and std::false_type above).

Failing at the substitution stage is different than general failure, and as it has to be an expression you are somewhat limited in what you can do. However, to test if copy is possible, you can do:

template<typename T, typename=void>struct copy_allowed:std::false_type{};
template<typename T>struct copy_allowed<T,
  sink_t<decltype(
    T( std::declval<T const&>() )
  )>
>:std::false_type {};

and move:

template<typename T, typename=void>struct move_allowed:std::false_type{};
template<typename T>struct move_allowed<T,
  sink_t<decltype(
    T( std::declval<T>() )
  )>
>:std::false_type {};

and only move:

template<typename T>struct only_move_allowed:
  std::integral_constant<bool, move_allowed<T>::value && !copy_allowed<T>::value >
{};

The general technique above relies on SFINAE. The base traits class looks like:

template<class T, typename=void> struct whatever:std::false_type{};

Here, we take a type T, and a second (anonymous) parameter we default to void. In an industrial strength library, we'd hide this as an implementation detail (the public trait would forward to this kind of private trait.

Then we specialize.

template<typename T>struct whatever<T, /*some type expression*/>:std::true_type{};

the trick is that we make /*some type expression*/ evaluate to the type void if and only if we want our test to pass. If it fails, we can either evaluate to a non-void type, or simply have substitution failure occur.

If and only if it evaluates to void do we get true_type.

The sink_t< some type expression> technique takes any type expression and turns it into void: basically it is a test for substitution failure. sink in graph theory refers to a place where things flow into, and nothing comes out of -- in this case, void is nothing, and the type flows into it.

For the type expression, we use decltype( some non-type expression ), which lets us evaluate it in a "fake" context where we just throw away the result. The non-type expression is now being evaluated only for the purpose of SFINAE.

Note that MSVC 2013 has limited or no support for this particular step. They call it "expression SFINAE". Alternative techniques have to be used.

The non-type expression gets its type evaluated. It isn't actually run, and it does not cause ODR usage of anything. So we can use std::declval<X>() to generate "fake" instances of a type X. We use X& for lvalues, X for rvalues, and X const& for const lvalues.