C++ implementation of strong types
Build Status | |
---|---|
Linux (gcc-8, clang-8) / OSX |
This tiny library provides a way to wrap an existing type in order to give it additional meaning. If you are already familiar with the concept of strong types, you can skip directly to the Examples section.
Otherwise, read along !
Let's take a basic example: we want to represent a distance in our code.
The immediate idea we could have would be to use an integral type, such as int
:
int distance_from_a_to_b = 10;
However, the type of the variable we work with does not convey any information about what its value actually represents. The only thing it tells us is how it is implemented. As a programmer reading the above code, you would need to rely on the name of the variable in order to understand the code.
This can be easily fixed using a type alias:
using distance = int;
distance from_a_to_b = 10;
That way, our code is more expressive, and easier to read for humans.
But we still have an issue ! From the compiler's point of view, the int
and distance
types are identical. This can lead to error-prone constructs:
int definitely_not_a_distance = 10;
distance from_a_to_b = definitely_not_a_distance; // Ouch !
The code above is not correct, because it allows converting definitely_not_a_distance
(which is clearly, not a distance) to a distance
object implicitly.
This is the first case for which this library can help: it can "hide" the real nature of a type, in order to prevent errors and unwanted conversions.
In order to fully hide the implementation of a type, we use the st::type
wrapper. It takes two (or more, but wait !) template parameters : the type to wrap, and a tag to guarantee its uniqueness.
using distance = st::type<int, struct distance_tag>;
Note: the tag can be any type, as long as it is only used as tag by a single strong type
Now, both the programmer and the compiler can distinguish a distance
from a regular int
.
int definitely_not_a_distance = 10;
int a_distance_value = 10;
// distance from_a_to_b = definitely_not_a_distance; // Not OK, would not compile
distance from_a_to_b = distance(a_distance); // OK
distance copy = from_a_to_b; // OK
As shown below, it is also possible to access the internal value of the strong type:
auto distance_value = from_a_to_b.value();
Now that we created a new type that hides its underlying implementation, we also lost access to the operations supported by the underlying type.
Why is that so ? Well, in our case, the concept represented by distance
might not support all the operations allowed by the int
type. For example, while you can add two distances together to make a longer distance, you clearly cannot multiply a distance with another distance. However, you can multiply a distance with a regular number, in order to scale it.
In order to customize the behavior of our strong types, this library uses the concept of traits. Traits are features that can be added to a type in order to give it additional behavior. Some basic traits are provided directly by the library (see the Built-In traits section), but it is also possible to write your own.
A strong type can use traits like below:
using distance = st::type<int, struct distance_tag,
st::addable, // distances can be added together
st::multiplicable_with<int> // distances can be scaled by a given factor
>;
This library provides two different ways to define strong types, each with different levels of complexity and flexibility.
This is the preferred way to create a basic strong type. It requires a type tag in order to guarantee the strength of the using
alias. Custom behavior can only be added through traits.
using integer = st::type<
int,
struct integer_tag,
st::arithmetic,
st::addable_with<int>
>;
This way makes it easier to customize a strong type because it skips the st::type
intermediate. Therefore, it requires creating a structure manually, which also allows defining custom member functions without having to use traits. However, traits are still available through inheritance.
struct int_with_a_member :
public st::type_base<int>,
public st::traits::arithmetic<int_with_a_member>
{
using st::type_base<int>::type_base;
constexpr bool is_zero() const noexcept
{
return value() == 0;
}
};
The table below describes the built-in traits that can be applied to a given strong type T
. Unless specified otherwise, these traits just forward the requested operation to the underlying types.
Trait | Behavior |
---|---|
addable |
Two T objects can be added to obtain a new T . |
addable_with<U> |
A T object can be added with a U object to obtain a new T . |
subtractable |
A T object can be subtracted from another T object to obtain a new T |
subtractable_to<U> |
A T object can be subtracted from another T object to obtain a new U . |
multiplicable |
Two T objects can be multiplied to obtain a new T . |
multiplicable_with<U> |
A T object can be multiplied with a U object to obtain a new T . |
dividable |
A T object can be divided by another T object to obtain a new T . |
modulable |
A T object can be moduled from another T object to obtain a new T . |
incrementable |
A T object can be pre-incremented and post-incremented. |
decrementable |
A T object can be pre-decremented and post-decremented. |
equality_comparable |
Two T objects can be compared for equality (supports == and != ). |
orderable |
Two T objects can be ordered (supports < , > , <= , >= ). |
arithmetic |
Shorthand trait for addable , subtractable , multiplicable , dividable , modulable , incrementable , decrementable , equality_comparable and orderable . |
bitwise_orable |
Two T objects can be bitwise OR -ed to obtain a new T . |
bitwise_orable_with<U> |
A T object can be bitwise OR -ed with a U object to obtain a new T . |
bitwise_andable |
Two T objects can be bitwise AND -ed to obtain a new T . |
bitwise_andable_with<U> |
A T object can be bitwise AND -ed with a U object to obtain a new T . |
bitwise_xorable |
Two T objects can be bitwise XOR -ed to obtain a new T . |
bitwise_xorable_with<U> |
A T object can be bitwise XOR -ed with a U object to obtain a new T . |
bitwise_negatable |
A T object can be bitwise negated (NOT ) to obtain a new T . |
bitwise_manipulable |
Shorthand trait for bitwise_orable ,bitwise_orable_with , bitwise_andable , bitwise_andable_with , bitwise_xorable , bitwise_xorable_with , bitwise_negatable and bitwise_manipulable . |