[Analysis / Pre-RFC] Variadic generics in Rust

[atagunov]

@yigal100

The idea is that a n-tuple should be a regular type that implements standard traits (like Iter ) instead of adding special unpack expression syntax

I fully agree with @PoignardAzur. The idea is to support a use-case Using `impl Trait` for monomorphized heterogenous collections

@yigal100

Unlike C++ which has a very bad track record in this regard

I understand part of bad track record is that C++ yields post-monomorphisation errors. E.g. errors that show up after a template has been instantiated. Rust has so far avoided such errors and hopefully would avoid them with ...T as well.

BTW post-monomorphisation errors is exactly where macros are worse than special language features like the one proposed here. Yes macros can do a lot of similar things but error messages would be less clear.

@yigal100

Well, I meant to use a tuple type constructor where each member is a type that implements the Sing trait.

I think

fn make_tuple_sing<T> (t: (T)) where T : Sing

syntax doesn't express this intent as clearly as either of

fn make_tuple_sing<(T)> (t: (T)) where T : Sing // hmm.. doesn't (T) mean something already?..
fn make_tuple_sing<...T> (t: ...T) where T : Sing
fn make_tuple_sing<...T> (...t: ...T) where T : Sing

@yigal100

The loop is executed by the compiler, not by your program

To me this is exactly the "a separate C++-like parameter pack ... syntax". These are templates akin to macros and akin to C++ but hopefully with much better error reporting and overall nicer to use. The only thing we're bike-shedding here is what exactly this syntax might look like.