Use sizeof(T) == 0 instead of always_false

[jmcarcell]

Introduced in #737 (also see #736 (comment)) I think it is not necessary to add that and the included header can be removed.

This is turbo pedantic but

Sorry :(

BEGINRELEASENOTES

• Use sizeof(T) == 0 instead of always_false

ENDRELEASENOTES

[m-fila]

In my defense P2593R1 (accepted for C++23) where relaying on sizeof(T) is explicitly listed as being wrong. But I agree in the end it doesn't really matter. We could also go back to sizeof(T) == 0

[jmcarcell]

Ah I didn't get that it was static_assert(false) what wasn't compiling, I thought it was sizeof(T) == 0

[m-fila]

Yeah, sizeof(T) && false, sizeof(T) == 0 and always_false are workarounds for static_assert(false) not behaving as desired. It's just that according to the standard always_false actually works, while the rest work by accident on most of the compilers

[tmadlener]

Given that the current always_false is a workaround for static_assert(false) only starting to work as expected in c++23. I would be in favor of keeping always_false because it's very clear in its name, whereas sizeof(T) == 0 needs some a bit more knowledge about why it is there.

[jmcarcell]

I don't have a strong opinion about this one. My preference would be not to include something that may or not be removed when switching to C++23 when there is already a (simple) way. So feel free to do whatever you want with this PR.

[m-fila]

My preference would be not to include something that may or not be removed when switching to C++23 when there is already a (simple) way.

Whichever workaround we choose now with C++23 we can replace it with static_assert(false)