Hacker News Reader

Comment by pjmlp | Hacker News Reader

The concept of unsafe keywords in systems programming languages goes back all the way to 1960.

Rust might have brought many interesting concepts into mainstream, however unsafe wasn't one of them.

That combined with marking types as Sendable points strongly at Rust influence. Then when you know that literal Graydon Hoare is working at Apple on Swift, the Rust influence is undeniable.

The Unsafe API has existed since Swift 1 which predates Graydon joining, and `UnsafeSendable` is just following the language naming conventions.

Perhaps this is exposing a gap in my knowledge of programming language history, but I'm not aware of other languages that mark types as threadsafe with an otherwise empty Send(able) implementation.

See Concurrent Haskell as possible set of ideas how typeclasses can be used to constrain the set of operations and sequences a type can perform.

But that's pretty different, since normal code in that context uses immutable data structures and the channels, etc. are language features. You then don't have to tack something like Send(able) onto arbitrary types because they're immutable anyway. The whole point of Send(able) is thread safety in the face of generally mutable state in contrast to something like Concurrent Haskell.

The idea is already there, the rest are details.

It is like arguing Rust invented affine types while ignoring the ideas in Cyclone and linear typing, even if they don't map 1:1 to what Rust has.

Affine types are a version of linear types.

Concurrent Haskell doesn't use typeclasses to constrain what data can be passed between threads. All data can be in Concurrent Haskell because all data is immutable anyway. It's a fundamentally different concept, not a slightly different subset like affine/linear types.

Sendable can also be equated to serializable once you are going cross-process. In that sense you see usage in systems like CORBA and RMI, and so systems that broadly aim to make things sendable like Erlang.

Is he still working there? It doesn’t really look like he is.

My sense is that was a recent change, a feature as integral to internal design f the language would be baking for while, and you would expect his influence to outlast his presence.

Graydon has been off doing other things for a while now. Off the top of my head, I think he left in 2019.

I miss working with him, but there are lots of folks involved with Swift who think really deeply about this stuff. No complex feature is the work of any one person in isolation.

Unsafe monads were already a thing in Haskell.

Sendable is just yet another monadic type, or a ML functor.

There is really no way in which Rust's Send or Swift's Sendable form a monad or a functor in any of the senses of those terms.

That doesn't change the fact that Haskell and ML did it first on their type systems, regardless how pedantic we want to discuss type theory.

The things you are saying are just not true. Standard ML and OCaml are in fact single-threaded languages, and while Haskell has concurrency and parallelism support it does not have anything that looks particularly like a Send or Sendable trait.

Concurrent ML:

http://cml.cs.uchicago.edu/pages/sync-var.html

As for the rest I am on the go to type a proper example.

Yes, Concurrent ML exists. It also, however, does not have any feature analogous to Rust's Send trait or Swift's Sendable protocol.

Of course it doesn't have a type class Sendable, that is not the point, rather the type theory behind it, and what those languages allow for, regardless of what they have in 2021 on their standard library.

While there is no exact match between them, the ``Send`` trait which can be thought of as preventing you from sharing an object between two threads actually has two monadic implementation in haskell. The ``IORef`` and ``STRef`` monads serve exactly the same purpose. This utilizes the Rank2Types extenstion. I invite you to read the Lazy Functional State Threads paper[1] by Simon Peyton Jones published in 1994.

The purpose of the ST Monad is to prevent you form sharing the object inside it wrapped by an STRef with another thread of execution.

IORef utilizes the uniqueness of the ``IO`` monad and its properties guaranteed by the haskell runtime i.e. only one thread can use it at any given time to prevent more than one thread from having a reference to the object inside the IORef from multiple.

[1]:https://www.microsoft.com/en-us/research/wp-content/uploads/...

Reply on news.ycombinator.com