## Rust 1.0 mistakes

Aaron’s List

NB: many of these are not “objectively” mistakes, but rather aturon’s opinion


- Bare `Trait` syntax (rather than `dyn Trait`)
- The `Error` trait (busted downcasting)
- The inconsistent privacy defaults with different data types
- The struct creation syntax (using `:`)
- Macro syntax future-proofing (which… isn’t)
- `io::Error` not working in `core`
- The `{:?}` syntax which is so annoying to type
- `Arc<F>` and `Rc<F>` not being `Fn(A) -> B` when `F` is
- Everything here: https://github.com/rust-lang/rust/issues?q=is%3Aissue+label%3Arust-2-breakage-wishlist+is%3Aclosed
- Including channels in `std`
- Exposing `String` repr such that the capacity hack for literals won’t work
- `PartialOrd`

Centril’s added opinion (implying agreement with the above set ^^):

- `struct` and `enum` should have just been `data`, i.e: one keyword
- type variables should have started with lowercase and concrete types with uppercase, and would permit `fn id(x: t) -> t` — tho this would lead to some possibly problems wrt. extern that you could overcome.
- the “head” type parameter of a trait is implicit

Niko’s thoughts for things beyond that set ( not implying agreement with all of them =) ):

- permitting “lifetime-sensitive dispatch” on traits
  - put another way: `#[``specialization_predicate``]` should have been the default
  - unlike most other things, can’t readily fix with an edition
- adding `AsRef` impls breaks type inference =)
- moving methods to inherent impls breaks type inference
- edition-sensitive method resolution
- `&uniq` — that prohibits mutation but guarantees uniqueness — might have been useful for things like `Pin`; I wonder if there is a way to rephrase `&mut T` in terms of `&uniq Mut<T>`, where `Mut` is analogous to `Cell` but requires an `&uniq` and not an `&T`.

Josh’s additions:

- `fmt::Write` and `io::Write` having the same name.

Simon:

- Range types implementing `Iterator` themselves, rather than just `IntoIterator`.

Yehuda’s additions:

- `mod` syntax
- such fine-grained privacy by default (vs. a more crate-centric default)
- String literals are not String
- No cross-crate coherence solution (some way to scope coherence to a bigger unit than “crate” that you still own)
- Controversial: `impl<T> Foo<T> {}`
  - I think in-band declaration should be allowed, just as it’s now allowed for lifetimes