Very interesting to see that Rust's "const generics" can emulate some of the flagship use cases of dependently typed programming such as having the length of an array be part of it's static type!
(source Wikipedia)
So rusts const generics are (very limited) dependent types.
C++ templates on the other hand you could argue are not types (but templates), but that is nit-picking.
Weather or not dependent types are can be runtime evaluated often is a separate question then weather or not a language has dependent types.
EDIT: The following part is partial nonsense, very POV dependent. I wanted to delete it but that wouldn't be right, so I just added this note.
Through not allowing run-time creation of dependent types makes them much more limited in usability. But then allowing it requires either reflections (still likely very limited/slow/etc.), or part of the type being stored as a "magic" field of the type instance (fast but is that still a dependent type?), or a interpreted language).
> The return type of a dependent function may depend on the value (not just type) of one of its arguments. For instance, a function that takes a positive integer n may return an array of length n, where the array length is part of the type of the array.
second:
> A dependent pair may have a second value of which the type depends on the first value. Sticking with the array example, a dependent pair may be used to pair an array with its length in a type-safe way.
Generally, it's the same nomenclature as with the explanation of static vs dynamic type systems: in static type systems, types are attributes of variables, functions and expressions, whereas in dynamic type systems types are attributes of values. And in the static type system of C, 5 has type int, because it's an expression. It representing a value is secondary. A side effect of it is: you can write a compile-time constant in C that will overflow its type, because the type is assigned based on the lexical category, not on the value.
As far as I understand, in a dependent type system it should be possible to implement a resizable array "std::vector<int, n>", where n is the length of the array (a variable), then write a "concat" function which concatenates two such arrays, and whose return (static) type depends on the lengths of the arrays passed as arguments. This is different than parametric polymorphism, such as the one that C++ has, since we can't know at compile-time all the possible lengths of arrays that we're going to get at runtime. Idris example here: https://idris2.readthedocs.io/en/latest/tutorial/typesfuns.h...
If someone has a better understanding of the topic, I'd be delighted to be shown wrong.
worksforme https://gcc.godbolt.org/z/x9GjE7KGM (ok, except the "resizeable" bit - but since we're doing functional programming we want immutable data structures, no ? :p)
Not only is it nitpicking, it seems wrong even in an academic sense of the term. Whether the template itself is a type or not probably depends on what theoretical language you use to discuss it.
The template is monomorphized into a new type per instantiation, yes, but the moment it actually has a number in it it's a type.
In D the template can quite happily accept pretty much anything as a parameter (particularly structs i.e. better code quality), and I'm surprised other languages don't allow this. Why have a bunch of template parameters on a line when you can just program the way you normally would.
Rust doesn't (arbitrary values, anyway, const generics is the start of this being possible) because we don't have templates, we have generics. I agree if you're going with a template-based system, it makes a lot of sense to accept whatever, but if you aren't, then things are more complicated.
* https://github.com/rust-lang/rfcs/pull/1657
* https://github.com/rust-lang/rfcs/issues/1930
* https://github.com/rust-lang/rfcs/pull/2000 (the design that was accepted, in the end)
(I am not actually 100% sure if rfc 2000 meets the by-the-book definition of "dependent types", my copy of Pierce has been collecting dust. Really gotta study up more on this stuff.)
One roundabout way of doing this - which is very much possible in D but not worth doing because you can't really do useful work with it for the most part due to pain - is that if you extend (say) const-generics to work with basically anything evaluable at compile time, you can build up a data structure to represent and hold true some predicate about a runtime value in a type (as a template/generic parameter), then define how these combine under binary operations (trivial so far in D at least, impossible in C++, no idea about Rust - monomorphization differences aside), then use this datastructure for evil and profit (like giving invariants to the optimizer).
I would like to see how far the aforementioned idea can go, however it quickly becomes "write a compiler at compile time" or worse "write an SMT solver at compile time", both of which do not thrill me.
fn make_array(n: usize) -> [u8; n]
I am guessing this isn't actually the feature though, and that it's not allowing dependence in this way, but rather promoting some const values and const functions to the type level (which is really useful, but not exactly dependent types)