[verdagon]

I can imagine a Rust-like language where we have move-constructors (in TFA), and every generated Future subtype is opaque and also heap allocated for us.

I think the need for Pin could then disappear, because the user would have no way to destroy it, since it's opaque and elsewhere on the heap, and therefore no way to move it (because having move-constructors implies that moving is conceptually destroying then recreating things).

[pornel]

Pin is a state rather than a property of the data itself.

This has a very nice effect of allowing merging and inlining of Futures before they're executed.

It's similar to how Rust does immutability — there's no immutable memory, only immutable references.

[withoutboats3]

You don't need move constructors if every future is heap allocated. But then every call to an async function is a separate allocation, which is terrible for memory locality. Some kind of virtual stack would be much better than that (but then you need garbage collection if you want the stacks to be memory-optimized to be small by default).

[anonymoushn]

You could use an actual stack. As I understand it this was not done for questionable reasons relating to borrows of thread-locals. You could also allocate a top-level async function and all of its transitive async callees all at once, if you force the user to put all this information in 1 translation unit. Or you could use a bump allocator specifically for futures used in a certain part of the program, if you're willing to give up using a global allocator for everything. So it seems like there are a lot of options.

[ComputerGuru]

All this is fine and dandy for async web apps but would make things like async-as-state-machine, which powers a lot of rust embedded development, basically a non-starter.

[anonymoushn]

In embedded development it's impossible to use any allocation strategy other than the language constantly calling a global malloc on your behalf?

[ComputerGuru]

Normal async doesn’t require any malloc.

[anonymoushn]

Great! I certainly didn't intend to say that stack-allocated or statically-allocated futures should somehow stop working?

[pcwalton]

If you're allocating a call stack per thread you might as well just use actual threads.

[anonymoushn]

This obviously doesn't work. Maybe you can try refactoring a console emulator that uses its stack switching primitive millions of times per second to use threads instead and let us know how well it runs.

[littlestymaar]

But you lose the async/await syntax (which I'm addicted to because it makes blocking point explicit, like `?` makes error explicit).

[josephg]

Yeah. I can guess how disruptive it would be, but I really wish rust bit the bullet and added a Move trait to std, baked into the language at a similar level as Copy. Move defines a function which moves a value from one address in memory to another. Structs without impl Move cannot be moved.

Almost all types would #[derive(Move)], which implements a trivial move function that copies the bytes. But this opens the door to self-referential types, futures, and lots of other things that need more complex move behaviour. (Actually, it might make more sense to define two traits, mirroring the difference between Copy and Clone. One is a marker trait which tells the compiler that the bytes can just be moved. The other allows custom "move constructor" implementation.)

I want move because pin is so hard to understand. Its a complex idea wrapped in double- or sometimes triple negatives. fn<X: !Unpin>(...). Wat? I drop off at unsafe pin-projecting. When is that safe? When is it not? Blah I'm out.

Moving from rust-without-move to rust-with-move would be inconvenient, because basically every struct anyone has written to date with rust needs #[derive(Move)] to be added. Including in std. And all types in existing editions that aren't pinned would need the compiler to infer a Move trait implementation. This should be possible to do mechanically. It would just be a lot of work.

Async rust is horrible. Especially compared to futures / promises in almost any other language. At some point, someone will make a new rust-like systems language which has an improved version of rust's memory safety model, a Move trait, and better futures. I'd personally also love comptime instead of rust's macro system.

I love rust. I love all the work the team has put into it over the years. But the language I’m really looking forward to is the language that comes after rust. Same idea, but something that has learned from rust’s mistakes. And it’s increasingly becoming clear what that better rust-like language might potentially look like. I can't wait.

[conradludgate]

I very much disagree that Pin is as hard as everyone makes it out to be. Using the pin! macros, the pin-project crate, and enough as_mut() to get it to compile and it's not hard at all to get a future impl working. It would be good to get this native (which is what boats wants) so it's easier to discover but it's not at all hard by any means

I think a lot of people think pin is confusing but don't actually try to learn it. When I've sat with people and helped them they understand pretty quickly what pin solves and how it works.

I very strongly think move constructors would be even more complex than pin.

[josephg]

I can only speak from my experience, but I really struggled with it. Ok, I understand moving. Pin is ... not moving. But the struct can still move, just ... not when you have a pointer to the struct. Ok, weird, but ok. Pin has a weird, limited set of functions to access the data fields of the struct. Some are unsafe. And then there's Unpin, which sounds like its not-not-move, so, something can move? No. From std:

> Implementing the Unpin trait for T expresses the fact that T is pinning-agnostic: it shall not expose nor rely on any pinning guarantees.

So, ??. Then there's macros for pin-project, which most projects in the wild use, but some are unsafe. Why? Which of my fields can I safely expose using pin-project?

I tried to implement a custom SSE-style streaming protocol over HTTP, to make a rust server implementation of Braid. I spent about a week trying, including pouring over the implementations of server-sent events and websockets in one of the HTTP libraries. Ultimately I failed to get it to work. (This is before TAIT and some other recent features, so things are probably be better now.)

I picked up javascript to write my server instead, and I had the protocol implemented about 20 minutes, in just 20 or so lines of code.

I adore rust, and I'd much rather a rust implementation than one based on nodejs. But I ran into skill issues here. Pin and futures in rust are hard. Or at least, I found them hard. I'm sure if I took another crack at it I'd be able to figure it out. But I don't want to spend so many of my brain cells on the language. I want to spend my attention thinking about my problem domain. Like I can in javascript.

Rust is an amazing language. But yeah, I really think that pin doesn't meet the standard that the rest of the language sets. I think it could use a rethink.

[withoutboats3]

I'm curious why you needed to deal with `Pin` instead of using async functions. What led you to a path in which you needed to implement poll methods yourself?

For what it's worth, all of the practical problems you encountered with using Pin are exactly what my next post is to show how to solve.

[josephg]

I look forward to your next post on the topic then!

> I'm curious why you needed to deal with `Pin` instead of using async functions.

The protocol I was trying to implement streams messages over time over a single HTTP request thats kept alive for a long time. This is how Server-Sent Events (SSE) works, and its how Google Chat in gmail was first implemented in a way that supported IE 5.5 (!!!).

This was a couple years ago now, so the details are a bit fuzzy. And I was relatively new to rust at the time. I was, at the time, still sometimes surprised by the borrow checker.

My goal was to make a writable async stream that I could push messages into from other parts of my program. And it also needed backpressure. When you sent messages into the stream, the protocol implementation it encoded them and streamed them into the body of my HTTP response object. I was (I think) using hyper.

This is before TAIT was in rust, and for one reason or another I needed to store / reference the future object I was making. (If you use an async fn(), you don't get a name for the Future type the function returns. So I couldn't put the return type in my struct, because I couldn't name it.)

So I ended up writing a custom struct that implemented Future, so I could reference the future elsewhere in my code. Hence, implementing Poll myself. I can't honestly remember how Pin came into it all. I think hyper's API for doing this sort of thing stored a Pin<T> or something.

I remember at some point trying to write a where clause using higher ranked type bounds to describe the lifetime of a future object that was- or wasn't- associated with the lifetime of the corresponding HTTP request. And that may or may not have been Pinned, and I gave up.

It might be fun to revisit this at some point now rust's async support has matured a little. And now that I've matured a lot in how I understand rust. I certainly don't imagine that everyone using async will run into the sort of quagmire that I hit. But this was the first thing I ever really wanted to do with async rust, and it felt horrible to fall on my face trying.

[withoutboats3]

Thanks for your write up. This sounds like a perfect use case for async generators (which yield many times and compile to Stream instead of Future), a feature I hope Rust will gain in the next year or two. To receive messages from other parts of the program, I would have the async generator hold the receiving end of a channel.

[raggi]

When I worked on Fuchsia and we had a heavily async set of system interfaces, users went through this learning path very regularly, and it was very painful for many. Folks who reached out for help early in their first engagement on this kind of path got help and following a "learn by doing" started to understand what was going on after a few iterations of the same challenge. Those who struggled trying to figure it out all on their own had a really awful time and in one example even went back to c++ for a sizable project because of the wall they ran into. There's a big gap here for folks who want to self-help their way through this. TAIT reduces the number of cases that come up, but there are still plenty.

Reflecting on a point from the article, it's possible that the ?Move being required in every declaration might have been better on this aspect. The point here about not being able to remember where the requirement to deal with Pin comes from is an indicator: the virality of the key traits involved, along with implicit implementation is a particularly tricky mix, it leads to action at a distance, which is also why first time engagements are so hard for users. Mix in some misunderstandings and you're in nope territory.

[sophacles]

> Async rust is horrible. Especially compared to futures / promises in almost any other language.

Having written a bunch of async code over 20 years (first exposure was twisted in the early 2000s and all sorts of stuff since - including manual function/stack wrangling on my own reactor), async in rust is like many other things in rust: It forces you to deal with the problems up-front rather than 6-12 months later once something is in prod. It helps to stop and understand why the compiler is yelling at you - for me anyway once I do grok that I'm glad I didn't have a repeat of $BUG_THAT_TOOK_DOWN_PROD_FOR_A_WEEK - that was a bad week.