DEV Community

That `overloaded` Trick: Overloading Lambdas in C++17

Tamir Bahar on August 15, 2017

C++17 has granted us with std::variant. Simply put, it is a type-safe union. To access the value it stores, you can either request a specific type ...

Read full post

Mikhail Matrosov • Jul 3 '18 • Edited

It would be nice to mention why we changed () to {} when we switched to deduction guide. E.g. if you add constructor

overloaded(Ts... ts) : Ts(ts)... {}

then you can use () again. At least I was curious about this.

Tamir Bahar • Jul 8 '18

I did mention that we need to use aggregate initialization. Though I now think my reasoning for it is wrong. Also - I never tried making it work with ().

Marcin Misiaszek • Apr 23 '19 • Edited

This also works without deduction guide:

template<class... Ts> struct overload : Ts... {
  overload(Ts...) = delete;
  using Ts::operator()...;
};

Ma-XX-oN • Apr 28 '19

I'm confused with this part:

        Print{ // (2)
            [](const char* str) { puts(str); },
            [](int i) { printf("%d\n", i); }
        }

The braces indicate a constructor call, but there is none. There is a non-member helper template though, but it has no body. So, what's happening here?

Ma-XX-oN • Apr 28 '19

It's not a non-member helper template, it's a user defined deduction guide. See this article:

arne-mertz.de/2017/06/class-templa...

Marcin Misiaszek • Apr 28 '19

It works also without deduction guide and the 'using' fix is for g++. 'operator()' cannot be ambiguous in visitors but gcc is very careful with overloading derived functions. In clang we can easily have the same results with minimal code (checked output assembler with Compiler Explorer):

template<class... Ts> struct overload : Ts... {
  overload(Ts...) = delete;
};

There is needed aggregate Initialisation + variadic templates.

Alexey Voinov • Aug 15 '17

Very cool code transformations. :) Thank you.

void78 • Mar 22 '18

Excellent brake-down, ta!

klaus triendl • May 28 '22

You don't need using Ts::operator()...;

Stanimir Lukanov • Oct 16 '17

That's just... weird...
PS: (...) this in the comment is not fold expression

Tamir Bahar • Oct 16 '17

What's weird?

Stanimir Lukanov • Oct 16 '17

Where the standard is going to :) nevertheless the implementation is cool!

Mikhail Matrosov • Jul 3 '18

(2) and (3) in second code snippet are mixed up.

Tamir Bahar • Jul 8 '18

Fixed. Thank you!