C++ C CMake
Clone or download
Latest commit 1a7762b Dec 4, 2017
Permalink
Failed to load latest commit information.
doc
test
tick
.travis.yml
CMakeLists.txt
LICENSE
README.md
appveyor.yml
biicode.conf
mkdocs.yml
tick.pc.in

README.md

Tick

Trait introspection and concept creator for C++11

Getting Started

Tick provides a mechanism for easily defining and using traits in C++11. For example, if we defined a generic increment function, like this:

template<class T>
void increment(T& x)
{
    x++;
}

If we pass something that does not have the ++ operator to increment, we will get an error inside of the increment function. This can make it unclear whether the error is due to a mistake by the user of the function or by the implementor of the function. Instead we want to check the type requirements of the function.

Using Tick we can create an is_incrementable trait, like this:

TICK_TRAIT(is_incrementable)
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(x++),
        decltype(++x)
    >;
};

And then we can use a simple requires clause in our function to check the type requirements:

template<class T, TICK_REQUIRES(is_incrementable<T>())>
void increment(T& x)
{
    x++;
}

So, now, if we pass something that is not incrementable to increment:

struct foo {};

foo f;
increment(f);

Then we get an error like this in clang:

demo.cpp:25:2: error: no matching function for call to 'increment'
        increment(f);
        ^~~~~~~~~
demo.cpp:14:19: note: candidate template ignored: disabled by 'enable_if' [with T = foo]
template<class T, TICK_REQUIRES(is_incrementable<T>())>
                  ^

This gives an error at the call to increment rather than inside the function, and then pointes to the type requirements of the function. This gives enough information for most commons cases, however, sometimes we may want more information. In that case the TICK_TRAIT_CHECK can be used. For example, say we had the is_incrementable trait defined like this:

TICK_TRAIT(is_incrementable, std::is_integral<_>)
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(x++),
        decltype(++x)
    >;
};

Then if we use TICK_TRAIT_CHECK, we can see why int* is not incrementable:

TICK_TRAIT_CHECK(is_incrementable<int*>);

Which will produce this error:

../tick/trait_check.h:95:38: error: implicit instantiation of undefined template 'tick::TRAIT_CHECK_FAILURE<std::is_integral<int *>, is_incrementable<int *> >'

Which shows the traits that failed including any refinements. So we can see that it failed because std::is_integral<int *> is not true.

Building traits

This macro will build a boolean type trait for you. Each trait has a require member function of the form:

TICK_TRAIT(my_trait)
{
    template<class T>
    auto require(T&& x) -> valid<
        ...
    >;
};

This will essentially build a class that inherits from integral_constant, so the above is equivalent to this:

template<class... Ts>
struct my_trait
: integral_constant<bool, (...)>
{};

The parameters to the trait are based on the parameters passed to the require function.

The trait will be either true or false if the expressions given are valid. Each expression in valid needs a decltype around it. If one of the expressions is not valid, the the trait will return false. For example,

TICK_TRAIT(my_trait)
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(x++)
    >;
};

The trait above will check that x++ is a valid expression.

Refinements

Refinements can be expressed after the name. Each refinement is a placeholder expression, where each placeholder(ie _1, _2, etc) are replaced by their corresponding type passed into the trait. In the case of traits that accept a single parameter the unnamed placeholder(_) can be used, for example:

TICK_TRAIT(is_incrementable, std::is_default_constructible<_>)
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(x++),
        decltype(++x)
    >;
};

This trait will be true when x++ and ++x are valid expressions and x is default constructible.

When a trait has multiple parameters, its best to use named placeholders. For example:

TICK_TRAIT(is_equality_comparable, 
    std::is_default_constructible<_1>, 
    std::is_default_constructible<_2>)
{
    template<class T, class U>
    auto require(T&& x, U&& y) -> valid<
        decltype(x == y),
        decltype(x != y)
    >;
};

This trait will be true when x == y and x != y are valid expressions and both x and y are default constructible.

In addition quote can be used to pass all the args from the trait to the refinement:

TICK_TRAIT(is_comparable, 
    quote<is_equality_comparable>)
{
    template<class T, class U>
    auto require(T&& x, U&& y) -> valid<
        decltype(x < y),
        decltype(x <= y),
        decltype(x >= y),
        decltype(x > y)
    >;
};

Query operations

These can be used to query more information about the types then just valid expressions.

Type matching

When a type is matched, it can either be convertible to the type given, or the evaluated placeholder expression must be true. Placeholder expressions can be given so the type can be matched against other traits.

returns

The returns query can check if the result of the expressions matches the type. For example,

TICK_TRAIT(is_incrementable)
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(returns<int>(x++))
    >;
};

This trait will be true if the expressions x++ is valid and is convertible to int.

Here's an example using placeholder expressions as well:

TICK_TRAIT(is_incrementable)
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(returns<std::is_integral<_>>(x++))
    >;
};

This trait will be true if the expressions x++ is valid and returns a type that is_integral.

Note: The TICK_RETURNS macro can be used instead to improve compatability with older compilers(such as gcc 4.6):

TICK_TRAIT(is_incrementable)
{
    template<class T>
    auto require(T&& x) -> valid<
        TICK_RETURNS(x++, int)
    >;
};

Also, returns<void> is prohibited.

TICK_TRAIT(is_incrementable)
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(returns<void>(x++)) // Compiler error
    >;
};

Instead, use either decltype directly without returns, or if there is a possibility of void from a computed type, use TICK_RETURNS or has_type instead.

has_type

The has_type query can check if a type exist and if the type matches. For example:

TICK_TRAIT(has_nested_type)
{
    template<class T>
    auto require(const T& x) -> valid<
        has_type<typename T::type>
    >;
};

This trait will be true if T has a nested type called type.

Now has_type used as above is not quite as useful since the above example, can also be simply written without has_type like this:

TICK_TRAIT(has_nested_type)
{
    template<class T>
    auto require(const T& x) -> valid<
        typename T::type
    >;
};

So, an optional second parameter can be provided to check if the type matches. Here's an example:

TICK_TRAIT(has_nested_int_type)
{
    template<class T>
    auto require(const T& x) -> valid<
        has_type<typename T::type, std::is_integral<_>>
    >;
};

This trait will be true if T has a nested type called type which is an integral type.

has_template

The has_template query can check if a template exist. For example:

TICK_TRAIT(has_nested_result)
{
    template<class T>
    auto require(const T& x) -> valid<
        has_template<T::template result>
    >;
};

This trait will be true if T has a nested template called result.

Trait evaluation

The is_true and is_false queries can check if a trait is true or false. Using refinements is the preferred way of checking for additional traits, but this can be useful if the evaluation of some trait can't be used lazily with placeholder expressions. So the is_true and is_false can be used instead, for example:

TICK_TRAIT(is_2d_array)
{
    template<class T>
    auto require(const T& x) -> valid<
        is_true<std::is_same<std::rank<T>::type, std::integral_constant<std::size_t, 2>> >
    >;
};

Helper functions

The library also provides as_const and as_mutable functions to ensure lvalues are either const or mutable respectively:

TICK_TRAIT(is_copy_assignable)
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(x = as_const(x))
    >;
};

Build traits without macros

The traits can be built without using the TICK_TRAIT macros. However, it may introduce problems with portability. So if only one platform is needed to be supported, then here's how to build them. First, build a class for the require functions and inherit from tick::ops to bring in all the query operations:

struct is_incrementable_r : tick::ops
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(x++),
        decltype(++x)
    >;
};

Next, turn it into a trait using tick::models:

template<class... Ts>
struct is_incrementable
: tick::models<is_incrementable_r, Ts...>
{};

Refinements

Refinements can be used by using the tick::refines class:

struct is_incrementable_r 
: tick::ops, tick::refines<std::is_default_constructible<tick::_>>
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(x++),
        decltype(++x)
    >;
};

Notice, the placeholders have to be fully qualified here.

Template constraints

Three macros are provided to help improve the readability of template constraints.

TICK_REQUIRES

The TICK_REQUIRES can be used on template parameters. For example,

template<class T, TICK_REQUIRES(is_incrementable<T>())>
void increment(T& x)
{
    x++;
}

TICK_CLASS_REQUIRES

The TICK_CLASS_REQUIRES can be used when template specialization is done on classes. For example,

template<class T, class=void>
struct foo
{
    ...
};

template<class T>
struct foo<T, TICK_CLASS_REQUIRES(is_incrementable<T>() and not std::is_integral<T>())>
{
    ...
};

template<class T>
struct foo<T, TICK_CLASS_REQUIRES(std::is_integral<T>())>
{
    ...
};

TICK_MEMBER_REQUIRES

The TICK_MEMBER_REQUIRES can be used for member function inside of classes, that are not templated. For example,

template<class T>
struct foo
{
    T x;

    TICK_MEMBER_REQUIRES(is_incrementable<T>())
    void up()
    {
        x++;
    }
};

TICK_PARAM_REQUIRES

The TICK_PARAM_REQUIRES can be used in the paramater of the function. This is useful for lambdas:

auto increment = [](auto& x, TICK_PARAM_REQUIRES(is_incrementable<decltype(x)>()))
{
    x++;
};

Also, the trait function is provided which can be used to deduce the type of the parameters:

auto increment = [](auto& x, TICK_PARAM_REQUIRES(trait<is_incrementable>(x)))
{
    x++;
};

Note: The trait function always deduces the type without references. So trait<std::is_lvalue_reference>(x) will always be false.

TICK_FUNCTION_REQUIRES

The TICK_FUNCTION_REQUIRES can be used on functions. This requires placing parenthesis around the return type:

template<class T>
TICK_FUNCTION_REQUIRES(is_incrementable<T>())
(void) increment(T& x)
{
    x++;
}

Note: The TICK_REQUIRES should be preferred.

Trait checking

The TICK_TRAIT_CHECK macro will statically assert the list of traits that are true but it will show what traits failed including base traits. This can be useful to show more informative messages about why a trait is false.

Requirements

This requires a C++11 compiler. There a no third-party dependencies. This has been tested on clang 3.4, gcc 4.6-4.9, and Visual Studio 2015.

ZLang support

ZLang is supported for some of the macros. The macros are in the tick namespace. For example,

$(trait is_incrementable)
{
    template<class T>
    auto require(T&& x) -> valid<
        decltype(x++),
        decltype(++x)
    >;
};

Acknowledgments