Re: [PATCH 00/45] C++: Convert the kernel to C++

[Alexey Dobriyan <adobriyan@gmail.com>]

On Thu, Jan 11, 2024 at 05:58:51AM -0500, Neal Gompa wrote:
> On Thu, Jan 11, 2024 at 5:56 AM Alexey Dobriyan <adobriyan@gmail.com> wrote:
> >
> > > SFINAE giving inscrutable errors is why I'm saying C++20,
> > > since "concept" means you can get usable error messages.
> >
> > I'd say concepts are irrelevant for the kernel where standard library is
> > tightly controlled by the same people who write rest of the kernel and
> > no external users.
> >
> > static_assert() is all you need.
> 
> We have external users all the time, though. People who write external
> modules or new modules would fall in that classification. Why should
> it be harder for them?

static_assert gives filename:line which clearly points to the source of
an error.

Concepts are SFINAE replacement but if there is little SFINAE there will
be little concepts too.

Another quite silly thing with concepts (and with noexcept propagation)
is that programmer has to write an implementation then chop control flow
and put everything else into the requires. And then to keep both in sync.

This is an example from cppreference:

	template<typename T>
		requires Addable<T>
	T
	add(T a, T b)
	{
		return a + b;
	}

Guess what Addable<> is?

They are kind of nice for simple things:

	template<typename It>
	concept MinimalIterator = requires(It it) { ++it; *it; it++; };