Implementing a Pipe Syntax for std::optional
- Part 2: Applied Template Metaprogramming
I’ve started implementing a pipe syntax to apply callables to optional values in a more expressive fashion here. However, I still needed a way to provide a useful return type if the callable itself returns an optional. I will implement this functionality using Template Metaprogramming.
Unraveling Optionals with Metafunctions
The puzzle piece I was missing in the last article was a metafunction that takes a type U
and returns a type depending on the given U
. Here, U
is the return type of the callable passed to the pipe operator. In the last article we saw that the metafunction needs to prevent nesting of optionals if the callable itself returns an optional type. The metafunction should behave like this:
- In case
U
is itself an optional, i.e.U
isstd::optional<V>
, return the typestd::optional<V>
. - In case
U
is void, give the compiler an illegal type that leads to a SFINAE when used as the return type of our operator1. - In all other cases: return the type
std::optional<U>
.
So in essence we have a switch case (or if-else) logic that we need to implement as a metafunction. This metafunction operates exclusively on types. That means that both the argument as well as the result of the function are types not values2. The way this is commonly implemented in C++ is to declare a struct
template with a typedef’d type
. The actual type of type
depends on the template specialization. Let’s see how exactly that works.
Implementing a Metafunction that Acts on Types
Our goal is to implement a simple metafunctions whose return type depends on a given argument type as described above.
Implementing the Default Case
I find it best to start with the default case, i.e. option 3 from above and do the special cases later:
#include <optional>
//default case of metafunction:
template<typename U>
struct wrap_return_type
{
typedef std::optional<U> type;
};
Our metafunction is implemented as a the structure wrap_return_type
. It takes a type U
and wraps it in a std::optional
. If we want to access the result for a given argument U
we have to write typename wrap_return_type<U>::type
. This is uneccessarily wordy and we can shorten it using a type alias:
template<typename U>
using wrap_return_type_t = typename wrap_return_type<U>::type;
I followed the naming convention of the standard by appending _t
to the name of my structure. This can now be used as a type by writing wrap_return_type_t<U>
, which is as legible as it gets for C++ Template Metaprogramming.
Implementing the Special Case for std::optional<U>
Let’s tackle case 1 from above, i.e. the special case that the argument type is already wrapped in an optional. We tackle this by providing a template specialization that is more specific than the general case:
template<typename V>
struct wrap_return_type<std::optional<V>>
{
typedef std::optional<V> type;
};
Now we just return exactly the optional type that was passed in without wrapping it in another optional3. Easy.
Implementing the Sfinae Case
Our goal above was to use wrap_return_type_t<U>
to find the return type of the pipe operator function. Here we use it to remove a particular overload, namely U
=void
, of the pipe operator. We do this by using SFINAE. Let’s do a quick recap of what that is: the beautiful acronym stands for Substitution Failure Is Not An Error:
[SFINAE] applies during overload resolution of function templates: When substituting the explicitly specified or deduced type for the template parameter fails, the specialization is discarded from the overload set instead of causing a compile error.
Since we are using the return type in a function template (the pipe operator), this really does apply here. If we want to remove a particular overload, we need to make template substitution fail. How do we do this? We just write an piece of code that would cause a compilation error in other circumstances. We can make wrap_return_type_t<void>
an invalid expression by not providing a type
typedef inside the template specialization:
template<>
struct wrap_return_type<void>
{
//type is not defined!
};
Since wrap_return_type_t<void>
is an alias for wrap_return_type<void>::type
this will lead to a substitution failure4. This substitution failure does not lead to a compilation error, but instead removes this particular overload from the set. That is exactly what we want.
The Pipe Operator with Unravelled Optionals
For the sake of completeness, let’s have a look at the implementation of the pipe operator now. Only one thing has changed compared to the last iteration, which is that the operator does not produce nested optionals anymore.
template<typename T, typename F>
auto operator|(const std::optional<T> & arg, F&& func)
-> wrap_return_type_t<std::invoke_result_t<F, T>>
{
if(arg.has_value())
{
return std::invoke(std::forward<F>(func), *arg);
}
else
{
return std::nullopt;
}
}
The rest of the implementation can be left unchanged, regardless of whether the callable returns an optional or not.
There is one last thing left to do. We removed this implementation for callables returning void
from the overload set using SFINAE. Now we have to provide a suitable overload instead. In the next article we will do just that.
Endnotes
-
I want to handle functions returning
void
with a different overload of the operator. That is why I want to use SFINAE to remove this implementation of the pipe operator from the overload set. Here is a very thorough introduction to SFINAE in Modern C++. I will also explain the magic behind this later in the article. ↩ -
In general, metafunctions in C++ can operate either on types or on compile time constant values (or both). The result of a metafunction, be it type or compile time constant, is always computed at compile time. ↩
-
I just named the template argument
V
instead ofU
to be consistend with the text above. I could just as well have named itU
. ↩ -
The way of provoking the substitution failre by removing the
type
from a template specialization is a standard solution. See e.g.std::enable_if
. ↩
Comments
You can comment on this post using your GitHub account.
Join the discussion for this article on this ticket. Comments appear on this page instantly.